Insurance policy complaint solution method, apparatus and device, and storage medium

Abstract

The invention belongs to the technical field of computer software, is applied to the field of medical health and the field of financial science and technology, and discloses an insurance policy complaint resolving method, device and equipment and a storage medium, and the method comprises the steps: obtaining customer data, judging a first state of a customer for a target insurance policy according to the customer data, the first state comprises a non-complaint state and a complaint state; constructing a first supervision model to obtain a complaint risk prediction result of the customer in the non-complaint state; implementing an interception strategy according to the complaint risk prediction result; constructing a second supervision model to obtain a solution difficulty level prediction result of the customer in the complaint state; and recommending a solution strategy according to the solution difficulty level prediction result. The invention provides an insurance policy complaint solution method and device, equipment and a storage medium, and solves the problems that the complaint risk cannot be predicted in advance and a solution strategy cannot be recommended according to the assessment solution difficulty in the prior art.

Description

Technical Field

[0001] The present invention belongs to the field of computer software technology and is applied to the fields of medical health and financial technology. In particular, it relates to a method, device, equipment and storage medium for resolving insurance policy complaints. Background Art

[0002] With the rapid development of the insurance industry, the number of customers and policy sizes continues to expand, leading to an increasing number of insurance policy complaints. These complaints not only impact customer satisfaction and loyalty but can also have a serious negative impact on an insurance company's reputation and business development. Traditional approaches to handling insurance policy complaints rely primarily on manual experience and post-processing, lacking systematicity and foresight.

[0003] When it comes to customer complaint risk management, existing methods often intervene passively only after a customer complaint occurs, failing to identify potential complaint risks in advance. Due to a lack of in-depth mining and analysis of customer data, it's difficult to accurately determine customer complaint tendencies, resulting in an inability to take effective preventative measures in a timely manner, increasing the probability of complaints. For complaints that have already occurred, traditional approaches are deficient in assessing the difficulty of resolution and recommending strategies. Complaint handlers often rely on personal experience to judge the difficulty of resolution and select handling strategies. Lacking scientific assessment models and data support, these methods result in inefficient handling, preventing some complaints from being resolved promptly and effectively, further exacerbating customer dissatisfaction and even triggering secondary complaints.

[0004] Therefore, existing technologies cannot predict complaint risks in advance, and cannot recommend resolution strategies based on the assessed difficulty of resolution. Summary of the Invention

[0005] The present invention provides a method, device, equipment and storage medium for resolving insurance policy complaints, so as to solve the problems existing in the video summary generation method in the prior art.

[0006] In a first aspect, the present invention provides a method for resolving insurance policy complaints, comprising:

[0007] Acquiring customer data, and determining a first status of the customer with respect to a target insurance policy based on the customer data, the first status including a non-complaint status and a complained status;

[0008] Constructing a first supervision model to predict the complaint risk of customers who have not filed a complaint, and obtaining a complaint risk prediction result for the customers who have not filed a complaint;

[0009] Implement interception strategies based on the complaint risk prediction results;

[0010] Constructing a second supervision model to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtaining a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint;

[0011] A resolution strategy is recommended based on the resolution difficulty level prediction result.

[0012] In a second aspect, the present invention provides a device for resolving insurance policy complaints, comprising:

[0013] an acquisition module, configured to acquire customer data and determine a first status of the customer with respect to a target insurance policy based on the customer data, wherein the first status includes a non-complaint status and a complained status;

[0014] A first prediction module is used to construct a first supervision model to predict the complaint risk of customers in the non-complaint state, and obtain a complaint risk prediction result of the customers in the non-complaint state;

[0015] An implementation module, used to implement an interception strategy based on the complaint risk prediction results;

[0016] A second prediction module is used to construct a second supervision model to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtain a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint;

[0017] The recommendation module is used to recommend a resolution strategy based on the prediction result of the resolution difficulty level.

[0018] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned method for resolving insurance policy complaints are implemented.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which implements the steps of the above-mentioned method for resolving insurance policy complaints when executed by a processor.

[0020] In the above-mentioned policy complaint resolution methods, devices, equipment and storage media, by constructing a first supervision model, high-complaint risk customers can be identified in advance, the influx of high-risk customers can be avoided, and the probability of complaints can be reduced from the source. By constructing a second supervision model, the attributes of complaints from customers who have already complained are analyzed, the difficulty of resolution is graded, and resolution strategies are dynamically recommended to avoid ordinary cases occupying high-quality resources, shorten the processing cycle, reduce repeated customer complaints, and improve the success rate of resolution. A digital model covering the entire process of "risk prediction-interception-resolution" reduces reliance on manual experience and improves the objectivity and consistency of decision-making. It is not only applicable to traditional insurance business, but can also be extended to cross-border scenarios such as healthcare, financial technology, etc. (such as health insurance, Internet insurance). Through cross-domain data fusion and model adaptation, the problem of complaint management in complex business scenarios can be effectively solved.

[0021] In summary, this solution can solve the problems that existing technologies cannot predict complaint risks in advance and cannot recommend resolution strategies based on the assessed resolution difficulty. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] Figure 1 This is a flow chart of a method for resolving insurance policy complaints in one embodiment of the present invention;

[0024] Figure 2 yes Figure 1 A flow chart of step S120;

[0025] Figure 3 yes Figure 2 A flow chart of step S130;

[0026] Figure 4 yes Figure 1 A flow chart of step S140;

[0027] Figure 5 yes Figure 1 A flow chart of step S150;

[0028] Figure 6 yes Figure 5 A flow chart of step S152;

[0029] Figure 7 yes Figure 5 A flow chart of step S153;

[0030] Figure 8 This is a structural diagram of a device for resolving insurance policy complaints in one embodiment of the present invention;

[0031] Figure 9 is a structural diagram of a computer device in one embodiment of the present invention;

[0032] Figure 10 FIG. 2 is another structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0034] See also Figure 1 As shown, an embodiment of the present invention provides a flowchart of a method for resolving insurance policy complaints, which includes the following steps.

[0035] Step S110 , obtaining customer data, and determining the first status of the customer with respect to the target insurance policy based on the customer data, wherein the first status includes a non-complaint status and a complaint status.

[0036] It should be noted that the customer data in step S110 is a comprehensive dataset covering multi-dimensional information, and its source and type are closely related to the specific business scenario. In the healthcare field, customer data is a carrier of health information related to insurance services, including but not limited to: customer health records (such as past medical history, family genetic diseases, and physical examination reports), medical behavior data (such as medical department, medical records, and length of hospital stay), and medical service evaluation (such as hospital ratings, doctor feedback, and complaint history). Chronic disease diagnosis results such as hypertension and diabetes are extracted from the customer's physical examination report and combined with historical medical frequency to form health risk assessment data. In the field of financial technology, customer data is a collection of data related to insurance policy transactions, credit assessment, or risk management, including: transaction flow (such as premium payment records, claim amounts, and investment returns), credit data (such as credit reports, default records, and credit scores), and operational behavior data (such as insurance operation time, cancellation frequency, and consultation records). The customer's transaction flow is analyzed to obtain the number and amount of premium payment delays, and the credit score is linked to generate risk warning indicators.

[0037] It should be noted that the customer data processing in step S110 uses multi-source data fusion and intelligent analysis technology. By establishing data cleaning, standardization and association rules, unstructured and semi-structured data are converted into structured data to support subsequent status judgment and model training. For example, in the medical and health field, natural language processing technology is used to extract key disease diagnosis information from the unstructured medical record text in the customer's health file and associate it with abnormal indicators in the physical examination data; in the field of financial technology, the customer's transaction flow and credit data are standardized, and combined with the high-frequency surrender operation records in the operation behavior data to construct a customer risk feature vector. Through these technical means, it is possible to accurately determine the customer's first status for the target insurance policy, providing a reliable data foundation for subsequent complaint risk prediction and resolution strategy formulation.

[0038] Step S120: construct a first supervision model to predict the complaint risk of the customers who have not filed a complaint, and obtain the complaint risk prediction results of the customers who have not filed a complaint.

[0039] It should be noted that the first supervisory model is a machine learning model used to predict customer complaint risk. The instructions for building this model can be triggered periodically by the insurance business system (e.g., risk prediction for newly insured customers at the beginning of each month) or in real time after a customer completes their insurance application. Customers with no complaints are those policyholders who have not yet filed a complaint. This model predicts the likelihood of future complaints, providing data support for insurance institutions to proactively intervene in services and optimize the customer experience.

[0040] In some embodiments of the present invention, Figure 2 As shown, step S120 includes the following steps:

[0041] Step S121, screening the target policies for policies covered within a specified time period;

[0042] Step S122 , defining the target variable as whether a complaint occurs within a preset time after the policy is underwritten;

[0043] Step S123: extracting static attributes and dynamic behavior characteristics of the customer before the insurance purchase, wherein the static attributes include the customer's occupation and age, and the dynamic behavior characteristics include the customer's historical complaint rate and channel negative feedback data;

[0044] Step S124: Divide the target policy into a training set and a validation set, and determine the model threshold based on the validation set performance;

[0045] Step S125: Record the customers whose probability is higher than the threshold as high-risk customers.

[0046] Specifically, in step S121, the method of screening the target policies that are insured within the specified time period can be any feasible method. For example, the database query function of the insurance core business system can be used to filter the policy data using SQL statements, and extract the policies that are insured throughout the specified year and are currently valid from the policy data table. The visual screening tool of the big data analysis platform can also be used to quickly locate the target policy by setting conditions such as time intervals and policy status. In the actual operation of the insurance business, there are differences in the screening logic for different types of insurance. In the field of health insurance, the focus is usually on screening the underwriting policies of newly launched products, because customers of new products may have deviations in their understanding of the terms and conditions, and the risk of complaints is higher; in the field of auto insurance, newly underwritten policies in specific areas (such as areas with a high incidence of traffic accidents) will be screened. Customers in these areas may be more likely to complain due to problems such as complicated claims processing.

[0047] Specifically, in step S122, the method of defining the target variable as whether a complaint occurs within a preset time after the policy is underwritten can be any feasible method. For example, based on the records of the insurance agency's customer service work order system, if a customer submits a valid complaint work order within 180 days after the policy is underwritten (determined by manual review or AI quality inspection to be a substantive complaint that is not a consultation or suggestion), the target variable corresponding to the customer is marked as 1; if no complaint work order is submitted, it is marked as 0. In order to define the target variable more accurately, it can be refined in combination with the complaint resolution situation. In the field of medical and health insurance, if a customer obtains full compensation after complaining, but is still dissatisfied with the claim settlement time limit and complains again, it can be marked as a high-priority complaint; in the field of financial insurance, complaints involving disputes over contract terms are marked as high-risk target variables to highlight their potential impact on the company's reputation and customer relationships. At the same time, natural language processing technology is used to perform semantic analysis on the complaint work order text to exclude invalid complaints and ensure the accuracy of the target variable definition.

[0048] Specifically, in step S123, the method for extracting static attributes and dynamic behavioral characteristics of pre-insurance customers can be any feasible method. For example, static attributes, such as the customer's occupation, age, gender, and income level, can be directly obtained from the information form filled out by the customer when applying for insurance and the insurance institution's customer information database. Dynamic behavioral characteristics can be extracted by analyzing historical interactions between the customer and the insurance institution. Log analysis tools can be used to count the number of times the customer called the customer service hotline and the main types of inquiries in the three months before the insurance application to assess the customer's service needs. App operation logs can be mined to obtain data such as the length of time the customer browsed the policy terms and the number of times the customer modified their insurance information. In the healthcare insurance sector, static attributes such as the customer's past health insurance claims history and whether they have chronic diseases can also be extracted, as well as dynamic behavioral characteristics such as whether they frequently consult about health services. In the financial insurance sector, static attributes such as the customer's credit rating and investment risk preference can be extracted, as well as dynamic behavioral characteristics such as whether they have repeatedly compared different insurance products. By combining deep learning with traditional statistical methods, the extracted features are subjected to dimensionality reduction and importance screening, retaining those that contribute most to complaint risk prediction.

[0049] Specifically, in step S124, the target policy is divided into a training set and a validation set. The method of determining the model threshold based on the performance of the validation set can be any feasible method. For example, a stratified random sampling method can be used to divide the target policy data into a training set and a validation set in a ratio of 7:3, ensuring that the ratio of positive and negative samples (complaining customers and non-complaining customers) in the two data sets is similar to avoid deviations in model training. A supervised learning model such as logistic regression and XGBoost is selected, and the model is trained using the training set, and then the trained model is evaluated using the validation set. By adjusting the classification threshold of the model and observing the changes in indicators such as precision, recall, F1 value, AUC-ROC on the validation set, the threshold that optimizes the overall performance is found. In the prediction of medical and health insurance complaint risk, if more emphasis is placed on reducing missed judgments (i.e., identifying all possible complaining customers as much as possible), a threshold with a higher recall rate can be selected; in the financial insurance field, if more attention is paid to reducing the waste of service resources caused by misjudgment, a threshold with a higher precision rate can be selected. At the same time, the cross-validation method is used to divide the training set and validation set multiple times for model training and evaluation, and the average performance index is taken to determine the final threshold to improve the reliability of threshold determination.

[0050] Specifically, in step S125, the method of marking customers with a probability higher than the threshold as high-risk customers can be any feasible method. For example, the trained first supervisory model is applied to customer data in a non-complaint state, and the model outputs the predicted probability of each customer making a complaint in the future. The probability is compared with the threshold determined in step S124. If the customer's predicted probability is greater than the threshold, such as the threshold is 0.6 and the predicted probability of a customer is 0.7, then the customer is marked as a high-risk customer and entered into the insurance institution's customer risk warning system. For customers marked as high-risk, insurance institutions can take targeted measures. In the field of medical and health insurance, high-risk customers are equipped with exclusive health consultants, who conduct regular health follow-up visits and interpretation of policy terms; in the field of financial insurance, high-risk customers are given enhanced secondary interpretation of contract terms and provided with personalized risk warning services. At the same time, the subsequent behavior of high-risk customers is continuously tracked, and feedback data is collected to further optimize the first supervisory model and improve the accuracy and effectiveness of complaint risk prediction.

[0051] It's understandable that the aforementioned series of steps for predicting complaint risk through the construction of the first supervisory model can more objectively and accurately identify potential high-complaint-risk customers than traditional methods that rely on experience to determine customer complaint risk. In the healthcare insurance sector, this can enable early intervention in services, reducing complaints stemming from a lack of understanding of health services and policy terms, thereby improving customer satisfaction and renewal rates. In the financial insurance sector, it helps optimize service resource allocation, reducing reputational risks and financial losses caused by customer complaints, providing strong support for insurance institutions' refined operations and risk prevention and control, and driving the development of the insurance business towards intelligence and efficiency.

[0052] Step S130: Implement an interception strategy based on the complaint risk prediction result.

[0053] It's important to note that the interception strategy aims to reduce the likelihood of subsequent complaints and alleviate pressure on customer service and business processing through targeted intervention for high-risk customers. This strategy can be automatically triggered by the system at a set time (such as every morning) or manually triggered when a specific high-risk event is detected (such as a large amount of negative feedback within a short period of time). Complaint risk prediction results are derived from the analysis and judgment of customer data by a previously constructed supervisory model.

[0054] In some embodiments of the present invention, Figure 3 As shown, step S130 includes the following steps:

[0055] Step S131, determining the second status of the customer with respect to the target insurance policy based on the customer data, where the second status includes an uninsured state and an insured state;

[0056] Step S132: construct a third supervision model to intercept the high-risk customers who are in the uninsured state from purchasing insurance, and guide the high-risk customers who are in the insured state to cancel their insurance.

[0057] Specifically, in step S131, the method for determining the second status of the customer for the target insurance policy based on the customer data can be any feasible method. For example, the judgment can be made by integrating multi-source data in the insurance business system. Obtain the customer's effective insurance policy record from the insurance policy management system, obtain the insurance information in transit from the insurance application system, and confirm the premium payment status from the payment system. Use SQL statements to query the data. If the query has results, the customer is in the insured state; if there are no results, and there is no corresponding insurance application record or the application has been rejected, it is determined to be uninsured. The real-time data processing capabilities of the big data analysis platform can also be used to continuously monitor changes in customer insurance-related data through a streaming computing engine (such as Flink) to dynamically update the customer status. In the health insurance business, customers who are in the waiting period after insurance are also considered to be insured; in the car insurance business, if the customer has completed the insurance but the vehicle has not yet been insured, it is also classified as insured for management.

[0058] Specifically, in step S132, a third supervisory model is constructed to intercept the uninsured high-risk customers from applying for insurance, and to guide the insured high-risk customers to cancel their insurance. Any feasible method can be used. For example, when constructing the third supervisory model, an ensemble learning algorithm (such as Random Forest or XGBoost) can be used to intercept the uninsured high-risk customers' insurance applications. This algorithm uses the customer's age, occupation, historical complaint history, and predicted probability of complaint risk as input features. Through training on historical data, it outputs the probability of the customer's insurance application being intercepted. When the predicted probability exceeds a set threshold (e.g., 0.7), an interception operation is implemented. This interception method can include a pop-up prompt on the insurance application page, informing the customer, "Based on our evaluation, the current product may not be suitable for you. We recommend contacting customer service (Phone: XXX) for a more suitable solution," directly blocking the insurance application process. Alternatively, an additional verification step can be added, such as requiring the customer to complete a policy-related knowledge questionnaire with an accuracy rate of at least 80% before continuing with the insurance application.

[0059] To guide high-risk customers to surrender their policies, a logistic regression-based model can be constructed. Input features include policy duration, premium amount, frequency of recent customer service communications, and the proportion of negative sentiment in these communications. The model outputs the probability of successful surrender guidance. For customers with a high probability (e.g., ≥0.6), customer service representatives will proactively initiate a phone call, stating, "We've noticed you have some questions about your policy. To ensure you receive the most appropriate coverage, we can review the policy details in detail or adjust your plan to your needs. If you still have concerns, we can also assist you with surrender procedures if regulations are met." Furthermore, a convenient surrender calculation tool is provided on the company's app or website, displaying the refundable amount and guiding customers to make their own surrender decisions. In the financial insurance sector, for high-net-worth, high-risk customers with existing policies, in addition to standard guidance, customized policy optimization solutions can be provided to reduce the likelihood of direct surrender. In the health insurance sector, if a customer becomes high-risk due to dissatisfaction with coverage, supplementary insurance products can be recommended to meet their needs and reduce surrender rates.

[0060] Specifically, during the implementation of interception strategies, to ensure their effectiveness and compliance, the entire process requires strict monitoring and management. A policy execution log system should be established to record in detail the time, customer information, operational rationale, model output, and other data for each interception or guidance action, facilitating subsequent audits and reviews. Third-party monitoring models should be regularly evaluated and optimized. By comparing actual complaints with model predictions, metrics such as accuracy, recall, and F1 score should be calculated. Model parameters should be adjusted or the model should be retrained based on the evaluation results. Furthermore, a policy adjustment mechanism should be established to promptly adjust the thresholds, methods, and language of interception and guidance strategies when market conditions, customer demographics, or company business policies change. In healthcare insurance, if new insurance regulations are issued that alter policy cancellation terms, the policy content for policy cancellation guidance should be updated promptly to ensure compliance with regulatory requirements. In financial insurance, if customer complaint hotspots shift, the criteria for identifying high-risk customers and interception strategies should be adjusted to address these new risk points.

[0061] It is understandable that the interception strategy implemented through step S130 can intervene in high-risk customers more proactively and accurately than the traditional passive complaint handling method. In the insurance business, it effectively reduces the number of complaints caused by customers misunderstanding policy terms, service mismatches and other issues, reduces the waste of customer service resources and the company's reputation risk. By intercepting uninsured high-risk customers, a large number of claims and complaint disputes that may arise in the future are avoided; guiding the cancellation of insured high-risk customers optimizes the policy structure to a certain extent and reduces the cost of compensation. The implementation of this strategy provides strong support for the insurance industry to achieve refined operations, improve customer satisfaction and business benefits, and promote the development of insurance business in a more intelligent and efficient direction.

[0062] Step S140 , constructing a second supervisory model, predicting the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtaining a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint.

[0063] It should be noted that the operation of the second supervisory model relies on multi-source data such as the customer service ticket system, customer relationship management system, and business database. The target variable is set as whether the customer complains again within a preset time after the complaint, in order to measure the difficulty of resolving the complaint. Model training supports both manual triggering (such as manual start after data update) and automatic triggering (such as scheduled training every morning). Customers who have filed a complaint refer to those who have submitted complaints through official channels such as telephone, official website, and APP, and whose tickets are in the "pending" or "processing" status.

[0064] In some embodiments of the present invention, Figure 4 As shown, step S140 includes the following steps:

[0065] Step S141, screening the customers who have filed complaints;

[0066] Step S142, defining the target variable as whether there is another complaint within a preset time after the complaint;

[0067] Step S143, extracting attributes and historical behavior correlation features of the customer's complaint;

[0068] Step S144: training a solution difficulty prediction model and outputting three difficulty levels of solution: high, medium, and low.

[0069] Specifically, in step S141, filtering out customers who have already filed a complaint can be accomplished in a variety of ways. Customer records in the complaint process can be filtered from the customer service ticket system based on the ticket status field (e.g., "pending" or "processing"). At the same time, the customer information table can be linked to the customer ID to obtain the customer's basic attributes (age, gender, region, etc.) and business attributes (insurance period, insurance type, etc.). The real-time data processing capabilities of a big data platform can also be utilized to perform real-time cleaning and filtering of accessed multi-source data to promptly identify eligible customers who have filed a complaint. A rule engine can also be used to set complex filtering rules, such as excluding records with a complaint type of "consultation" and a ticket processing time of less than one hour, to accurately locate valid complaining customers. In the financial insurance sector, when screening auto insurance complaining customers, high-profile complaint cases can be further screened based on the accident type (e.g., multi-party accidents, major personal injury accidents). In the health insurance sector, complaints involving major disease claims disputes can be specifically screened.

[0070] Specifically, in step S142, the target variable is defined as whether a complaint is filed again within a preset timeframe after the complaint. The timeframe and criteria for determining if a complaint is filed again require clear definition. The timeframe can be flexibly set based on business characteristics, such as 30 days for auto insurance complaints and 60 days for health insurance complaints. To determine if a complaint is a repeat, natural language processing techniques can be used to analyze the semantic similarity of the complaint text. If the cosine similarity between the new complaint text and the previous complaint text exceeds 0.7 (the threshold is adjustable) and occurs within a preset timeframe, it is considered a repeat complaint. Alternatively, a comprehensive assessment can be made based on information such as the complaint type and the policy involved. In practice, multiple complaints from the same customer involving different policies or different types of issues are treated as independent events and marked separately. In e-commerce insurance scenarios, even if customer claims for different orders fall within a specific timeframe, it is necessary to determine whether they are repeat complaints based on the specific circumstances. In the corporate group insurance sector, multiple employee complaints regarding different issues related to group health insurance also require detailed analysis.

[0071] Specifically, in step S143, the extraction of attributes and historical behavior-related features of the customer's complaints can be carried out from multiple dimensions. In terms of customer attributes, it covers basic information (age, occupation, spending power, etc.), policy information (premium amount, coverage period, claim record, etc.); in terms of historical behavior, it includes the number of historical complaints, complaint intervals, complaint channel preferences, complaint handling satisfaction, etc. Data mining technology can be used to extract these features from massive historical data. For example, clustering algorithms can be used to analyze customer complaint behavior patterns and identify characteristic portraits of high-frequency complaint customer groups. Feature engineering methods in machine learning are used to normalize and encode the original data to enhance the expressive power of features. In the bank credit card insurance business, features such as customer credit scores and card usage activity are extracted; in the field of travel insurance, features such as customer travel frequency and destination risk level are extracted for complaint difficulty prediction.

[0072] Specifically, in step S144, a resolution difficulty prediction model is trained to output difficulty labels of high, medium, and low difficulty levels. This can be achieved using a variety of machine learning algorithms. Common traditional algorithms include decision trees, random forests, support vector machines, and neural network models in deep learning. Taking the random forest algorithm as an example, multiple decision tree models are constructed and trained on the input customer feature data. Ultimately, the customer's resolution difficulty level is determined through a voting mechanism. During the training process, model parameters can be optimized through methods such as cross-validation and grid search to improve prediction accuracy. After model training is complete, the probability of each customer belonging to the high, medium, or low difficulty level is output. The final label is determined based on a preset threshold (e.g., high difficulty probability ≥ 0.7, low difficulty probability ≤ 0.3). In actual applications, the model can be regularly updated and optimized according to business needs, for example, retraining the model monthly to adapt to changes in customer behavior and the business environment. In the internet insurance scenario, the rapidly iterative business model requires the model to have strong adaptability and timely parameter adjustment to ensure prediction accuracy. In traditional insurance institutions, the model is optimized based on new business policies and customer feedback.

[0073] It is understandable that by constructing a second supervisory model to predict the difficulty level of resolving complaints from customers who have already filed a complaint, the difficulty of complaint handling can be assessed more comprehensively and objectively compared to the traditional manual experience judgment method. In the financial insurance industry, accurate difficulty predictions help customer service teams to allocate resources in advance, give priority to high-difficulty complaints, and avoid escalating customer dissatisfaction. By learning from a large amount of historical data, the model can discover the laws and patterns hidden behind the data and provide data support for companies to formulate customer service strategies. For example, for customer groups with high frequency and high difficulty complaints, companies can optimize product design, improve service processes, and reduce complaints from the root; for customers with low difficulty complaints, they can improve processing efficiency and reduce operating costs through automated service processes, thereby achieving refined management of customer service and efficient use of corporate resources.

[0074] Step S150: recommending a resolution strategy based on the resolution difficulty level prediction result.

[0075] This step is triggered based on the difficulty level (high, medium, or low) output in step S140. Recommended resolution strategies include personnel assignment, recommended conversational techniques, and decision-making regarding settlement funds. Strategy recommendations can be automatically generated by the system or manually retrieved and adjusted by customer service personnel based on actual circumstances. This system is applicable to customer complaint handling scenarios across multiple industries, including insurance, e-commerce, and finance.

[0076] In some embodiments of the present invention, Figure 5 As shown, step S150 includes the following steps:

[0077] Step S151, assigning the customer who has filed a complaint to a primary, intermediate, or expert level resolution personnel according to the resolution difficulty level;

[0078] Step S152: recommending a speech strategy with the highest success rate to the resolution personnel based on the characteristics of the customer who has filed a complaint and historical resolution data;

[0079] Step S153: For customers with high-difficulty complaints, the assisted settlement model is used to determine whether assisted settlement funds are needed, and experts are assigned or assisted settlement fund plans are initiated based on the results.

[0080] Specifically, in step S151, different levels of resolution personnel are assigned based on the difficulty level of the resolution. This can be achieved in a variety of ways. A database can be established that maps personnel skills to difficulty levels. Junior resolution personnel are assigned to handle low-difficulty complaints, typically simple consultations and basic business processes, such as policy information inquiries and simple claim submission guidance. These require junior personnel to be familiar with basic business knowledge and operational procedures. Intermediate resolution personnel handle medium-difficulty complaints, such as general claims disputes and disputes over understanding terms and conditions. These require intermediate personnel to possess certain case analysis and communication and coordination skills. Expert resolution personnel are assigned to handle high-difficulty complaints, such as major claim denial disputes and mass complaint incidents. These require experts to possess deep professional knowledge, extensive handling experience, and cross-departmental coordination capabilities. When assigning resolution personnel, an intelligent scheduling algorithm is used to comprehensively consider factors such as the resolution personnel's current workload, professional skills, historical processing efficiency, and satisfaction. For example, a load balancing algorithm can prioritize the allocation of complaints to personnel with the least number of currently processed work orders and the appropriate skills. For high-difficulty complaints, an expert priority principle can be established to automatically assign them to expert personnel with the highest historical success rate and the most experience in handling these complaints. In the insurance industry, auto insurance claims complaints involving minor scratches (low difficulty) are assigned to junior staff; complaints involving multiple parties with unclear liability determinations (medium difficulty) are assigned to mid-level staff; and complaints involving major personal injury accidents with complex liability disputes (high difficulty) are directly assigned to expert staff who specialize in handling such cases. In the e-commerce industry, inquiries about product return and exchange procedures (low difficulty) are handled by junior customer service representatives, product quality disputes (medium difficulty) are followed up by mid-level customer service representatives, and consumer disputes involving large compensation amounts (high difficulty) are handled by expert customer service representatives.

[0081] Specifically, in step S152, a recommended speech strategy is based on customer characteristics and historical resolution data, using a variety of technical approaches. Natural language processing (NLP) technology is used to analyze historical cases of successful resolution, extract effective speech templates, and classify and store them according to complaint type and customer group characteristics to build a speech library. In practice, when a new complaint is received, the system first analyzes the customer's basic characteristics, such as age, gender, consumption habits, and historical complaint records. Key information about the current complaint, such as complaint type, reason, and customer sentiment, is also extracted. Then, using text similarity algorithms such as cosine similarity and edit distance, the speech library is searched for historical cases and corresponding speech templates that are most similar to the current complaint scenario. For example, if the customer is elderly and has filed a complaint about product usage, the speech library is searched for product usage guidance scripts tailored to elderly customers. If the customer is emotionally agitated, speech templates that soothe and resolve the issue are prioritized. Deep learning models, such as language models based on the Transformer architecture, can also be combined to dynamically generate follow-up speech based on the customer's real-time responses, making communication smoother and more tailored to customer needs. In the financial industry, for complaints about dissatisfaction with financial product returns, we recommend different sales pitches based on the customer's risk tolerance, investment experience, and other characteristics. For conservative customers, we emphasize risk control and stable returns, while for experienced customers, we explain the market situation and product strategy. In the telecommunications industry, for complaints about poor network signals, we recommend sales pitches that include troubleshooting methods and compensation plans based on the customer's location and investment plan.

[0082] Specifically, in step S153, a settlement model is used to determine whether settlement funds will be used for high-difficulty complaint clients. Various algorithms and models can be used. The settlement decision model is constructed using input data including basic complaint case information, such as the complaint amount, dispute focus, and the terms of service involved; customer-related information, such as customer value rating, historical complaint history, and credit score; and historical data on the resolution outcomes of similar complaints. This data is trained using machine learning algorithms, such as random forests, support vector machines, or deep learning algorithms, such as neural networks, to learn the relationship between different factors and the inclusion of settlement funds. In actual judgments, the settlement model is fed with relevant data from high-difficulty complaint clients, and the model outputs a probability of settlement funds being involved or a recommended outcome. If the probability exceeds the set threshold (e.g., 70%), it is determined that a settlement fund is needed. The system will automatically initiate the settlement fund application approval process and assign expert personnel with extensive negotiation experience and financial knowledge to follow up. Based on the settlement amount range recommended by the model and the actual situation, the final settlement plan will be determined through negotiation with the customer. If the probability is lower than the threshold, the settlement fund plan will not be initiated, and the expert personnel will continue to resolve the complaint through other means. In insurance claims complaints, for cases involving large compensation and large disputes, the settlement model comprehensively considers factors such as the insured's medical expense details, insurance terms and conditions, and past settlement situations of similar cases to make judgments. In consumer disputes on e-commerce platforms, for complaints about quality disputes of high-value goods, the model refers to data such as product prices, customer purchase records, and platform processing rules to make decision recommendations.

[0083] It's understandable that by recommending resolution strategies based on predicted difficulty levels, compared to traditional manual judgment and handling methods, complaint handling can be more scientific and refined. In multi-industry applications, complaints can be quickly and accurately matched with appropriate handling resources and strategies, reducing processing time and wasted resources. By learning and analyzing historical data, recommended speech strategies and collaborative resolution decisions are more targeted and effective, helping to improve customer problem resolution rates and enhance customer trust and satisfaction with corporate services. At the same time, this intelligent recommendation mechanism can also provide data support for companies to accumulate handling experience and optimize business processes, promoting the continuous improvement of corporate service quality and management level, and gaining an advantage in market competition.

[0084] In some embodiments of the present invention, Figure 6 As shown, step S152 includes the following steps:

[0085] Step S1521, analyzing the success rate of historical sales pitches under different customer characteristics;

[0086] Step S1522: Match the optimal sales pitch based on the current customer's gender, age, and complaint channel characteristics;

[0087] Step S1523, dynamically adjust the priority of recommended speech skills based on the skill tags of the resolution personnel.

[0088] Step S1524: predict whether settlement funds are needed for high-difficulty complaint customers;

[0089] Step S1525: If it is predicted that no settlement is needed, expert personnel are assigned to handle the matter; if settlement is needed, a settlement fee recommendation is generated and the payment process is initiated.

[0090] Specifically, in step S1521, the success rate of historical speech under different customer characteristics can be analyzed in a variety of ways. First, it is necessary to collect and organize historical complaint processing data, including the basic characteristics of customers (gender, age, region, consumption level, etc.), the specific content of the complaint, the speech text used, the processing results (successfully resolved, unresolved, upgraded processing, etc.), and customer satisfaction evaluation information. Then, data mining and statistical analysis methods are used to clean and preprocess the data to remove noise and outliers. Classification algorithms such as decision trees and random forests can be used to build a success rate prediction model with customer characteristics and speech content as input variables and processing results as output variables. For example, using a random forest algorithm, by training a large number of decision trees, the success patterns of various speech under different customer feature combinations can be learned. In the financial insurance industry, analysis of the customer group "over 60 years old, from third-tier cities, and with a lower consumption level" found that using "plain language to explain insurance terms and emphasize benefits" has a higher success rate in handling insurance claims complaints; in the e-commerce industry, for "young women, high-frequency consumers, and those who complain through apps," using "quick apologies + providing exclusive compensation plans" is more likely to successfully resolve product quality complaints.

[0091] Specifically, in step S1522, the optimal speech is matched according to the gender, age, and complaint channel characteristics of the current customer, involving multiple technologies and strategies. First, natural language processing technology is used to analyze the current customer's complaint text, extract key information, and combine the customer's basic information (gender, age) and complaint channel (telephone, APP, web form, social media, etc.) to form a customer feature vector. Then, in the established speech library, text similarity calculations (such as cosine similarity, edit distance and other algorithms) are used to retrieve the speech corresponding to the historical success case with the highest matching degree with the current customer's feature vector. For example, if the customer is a young male and complains about the use of electronic products through the APP, the system will give priority to filtering out speech for young male users, APP channels, electronic product complaint scenarios and with a high historical success rate from the speech library, such as "Dear, don't worry! We do have some tips for operating this product. I will send you a super detailed video tutorial. You can solve the problem immediately by following it~". At the same time, the recommended language can be dynamically adjusted based on the customer's real-time feedback and emotional state. If it is detected that the customer is emotionally agitated, it will be promptly switched to language that focuses on soothing the emotions, and the problem can be solved after the customer's emotions are stabilized.

[0092] Specifically, in step S1523, the priority of recommended speech scripts is dynamically adjusted based on the skill tags of the dispute resolution personnel. This requires the establishment of a comprehensive personnel skill tag system and intelligent adjustment algorithm. Each dispute resolution personnel is assigned a skill tag, such as "Insurance Clause Interpretation Expert," "E-commerce After-Sales Dispute Mediation Expert," or "Customer Emotional Soothing Specialist." The proficiency of each skill is evaluated based on their historical complaint handling performance (such as handling success rate, customer satisfaction, and handling time). When recommending speech scripts for a dispute resolution personnel, the match between the speech script and the personnel's skill tags is calculated, and speech scripts with high matching scores are prioritized. For example, for a dispute resolution personnel with the "Insurance Clause Interpretation Expert" tag and high proficiency, when handling insurance clause dispute complaints, speech scripts involving detailed explanations of the clauses and citations of legal basis will be prioritized and displayed higher. For a dispute resolution personnel with the "Customer Emotional Soothing Specialist" skill tag, soothing and empathetic speech scripts will be ranked higher in the recommended list when handling complaints from emotionally agitated customers. Furthermore, the system supports manual intervention to adjust the recommendation order. Dispute resolution personnel can manually select appropriate speech scripts based on their actual situation. The system records these adjustments and uses them to optimize the recommendation model later.

[0093] Specifically, in step S1524, the need for settlement payment intervention is predicted for high-difficulty complainants using machine learning prediction models and business rules. Criteria for determining high-difficulty complainants are first determined, such as whether the complaint amount exceeds a certain threshold, the dispute is complex and involves multiple parties, or the customer has filed multiple unresolved complaints. Then, relevant data on high-difficulty complaints is collected, including the complaint amount, dispute focus, customer complaint history, customer credit score, and the company's handling of similar cases, as input features for the model. Machine learning algorithms, such as logistic regression, support vector machines, and neural networks, are used to train a binary classification model and output the probability of settlement payment intervention. For example, neural networks, through multi-layer learning and training, uncover the potential relationship between data features and the need for settlement payment. In practice, when encountering a high-difficulty complainant, the model inputs relevant data. If the predicted probability exceeds a set threshold (e.g., 0.6), settlement payment intervention is determined to be necessary; otherwise, it is not. In the insurance industry, for major disease insurance claim disputes involving high-value compensation, the model will make predictions based on a comprehensive consideration of factors such as the severity of the insured's condition, insurance terms and conditions, and past settlements of similar cases. In the e-commerce industry, for quality dispute complaints involving high-value goods, the model will make judgments based on data such as product prices, customer purchase records, and platform processing rules.

[0094] Specifically, in step S1525, if the prediction indicates no settlement is necessary, expert personnel are assigned to handle the case. If settlement is necessary, a recommended settlement fee is generated and the payment process is initiated. This requires the development of clear implementation rules and automated processes. If the prediction indicates no settlement fee is necessary, the system assigns the complaint to the most appropriate expert based on their expertise and skill sets to ensure a proper resolution. For example, a complaint involving a complex insurance contract dispute would be assigned to an expert familiar with insurance law and contract details. If a settlement fee is predicted to be necessary, a recommended settlement fee is first generated using a regression model or statistical methods based on historical data. For example, a reasonable settlement fee amount is calculated based on factors such as the dispute amount, the importance of the client, and the settlement fee ratios in similar cases. The automated payment process then begins, including submission of the settlement fee application, review by the finance department, negotiation and confirmation with the client, and final payment. Throughout the entire process, the system records every step and related information for subsequent review and audit. In the financial industry, the customer's identity and the authenticity of the disputed materials must be strictly reviewed before the settlement payment is made; in the e-commerce industry, settlement compensation may be completed through a combination of coupons, refunds, and other methods, while ensuring that the process complies with the platform's financial and operational regulations.

[0095] It is understandable that by recommending speech strategies based on the characteristics of customers who have already filed complaints and historical resolution data, compared to the traditional application of empirical speech techniques, scientific and precise complaint handling has been achieved. In multiple industry scenarios, it can quickly and accurately provide adaptive communication strategies for resolution personnel, effectively improving complaint handling efficiency and reducing customer waiting time and communication costs. Through in-depth analysis of historical data and model training, recommendation strategies are continuously optimized, the success rate of problem resolution is improved, and customer trust and satisfaction with corporate services are enhanced. At the same time, this mechanism can also help companies accumulate valuable customer communication and problem-solving experience, provide data support for business process optimization and product and service improvements, and help companies improve service quality and core competitiveness in market competition.

[0096] In some embodiments of the present invention, Figure 7 As shown, step S153 includes the following steps:

[0097] Step S1531: predict whether settlement funds are needed for high-difficulty complaint customers;

[0098] Step S1532: If it is predicted that no settlement is required, expert personnel are assigned to handle the matter; if settlement is required, a settlement fee recommendation is generated and the payment process is initiated.

[0099] Specifically, in step S1531, predicting whether a customer with a high-difficulty complaint requires a settlement payment can be achieved through a variety of methods. First, a comprehensive dataset must be constructed for model training and prediction. This dataset includes basic customer information, such as age, gender, and consumer level; complaint information, including complaint type (e.g., product quality issues, unfulfilled service promises), complaint duration, number of complaints, and the amount involved; and historical company processing data, such as the outcome of similar complaints, whether a settlement payment was used, and the final customer satisfaction rating. Next, a suitable machine learning algorithm can be selected to construct a settlement model, such as logistic regression, random forest, or gradient boosting tree algorithms. For example, the gradient boosting tree algorithm can continuously optimize the model through iterative learning, uncovering hidden patterns and relationships in the data, thereby probabilistically predicting whether a customer with a high-difficulty complaint requires a settlement payment. In practical applications, natural language processing techniques can also be combined to analyze complaint texts to extract keywords and sentiment as supplementary model features. For example, if a complaint text frequently contains strong expressions such as "must compensate" and "serious losses," the predicted probability of a settlement payment can be appropriately increased. The focus of prediction varies across different industries. In the e-commerce industry, if a high-difficulty complaint involves high-value goods and the customer is a high-frequency consumer, the model will comprehensively consider the customer's potential value and increase the possibility of mediation fund intervention; in the financial industry, for complaints involving investment losses and the customer has good credit, it will also tend to predict the intervention of mediation funds to maintain customer relationships and corporate reputation.

[0100] Specifically, in step S1532, if it is predicted that no settlement is necessary, an expert is assigned to handle the case. If settlement is necessary, a recommended settlement fee is generated and the payment process is initiated. The specific implementation is as follows. If settlement is not predicted, the system accurately matches the expert's professional skills with the complaint type. The company pre-establishes a detailed skill tag system for experts. For example, in the field of electronic product complaint handling, there may be experts specializing in hardware failure analysis and those skilled in software problem solving. An algorithm calculates the match between the complaint type and the expert's skill tag, and the complaint is assigned to the expert with the highest match and who is not currently overloaded, ensuring a professional and efficient resolution. If settlement is predicted to be necessary, a recommended settlement fee is first generated. This can be done using statistical methods based on historical data to retrieve settlement fee payments for similar complaints, taking the average or median as the baseline recommendation. Alternatively, a machine learning regression model can be used to input complaint-related features (such as the amount involved and customer importance indicators) to predict the settlement fee amount. After generating the recommended value, the payment process begins. This process is subject to a strict review process, including a review by the finance department for the reasonableness of the amount and a review by the legal department for the compliance of the agreement. Once approved, the settlement payment is completed through the company's payment system, and an electronic agreement is signed with the customer, clarifying the rights and obligations of both parties. In the financial industry, the settlement payment process must also comply with regulatory regulations on capital flow. In the service industry, customer satisfaction surveys are often conducted after payment is completed to collect feedback and optimize subsequent processing.

[0101] It's understandable that using a assisted resolution model to identify clients with high-difficulty complaints requiring assisted resolution funding and implementing appropriate measures is more accurate and efficient than traditional manual judgment and processing methods. This approach can rapidly process large amounts of complex complaint data, avoiding the subjectivity and limitations of manual decision-making. In financial services, it can accurately identify clients requiring assisted resolution funding, promptly resolve disputes, maintain customer trust, and reduce the risk of churn. In e-commerce retail, expert resources can be rationally allocated to handle complaints that don't require assisted resolution, accelerating customer resolution while scientifically determining assisted resolution funding amounts and balancing enterprise costs with customer satisfaction. Furthermore, this mechanism allows for the continuous accumulation of experience and optimization of the assisted resolution model through continuous learning from historical data, further enhancing the intelligence of complaint handling and the competitiveness of enterprise services.

[0102] In this way, the overall solution of the embodiment of the present invention realizes the closed-loop processing of "accurate prediction - active interception - intelligent solution" for customer complaints in industries such as insurance, finance, and medical care through a complaint management system driven by full-process data intelligence. First, through the multi-dimensional data integration and status recognition of step S110, multi-source data such as customer health records, transaction flows, and complaint history are collected, and a standardized data set is constructed in combination with data enhancement algorithms (such as simulated handwritten insurance policies and high-risk transaction scenarios) to accurately distinguish between customers' "no complaints" and "complaints" status. For example, in the health insurance scenario, combined with medical records, potential complaint risk customers can be identified in advance; in the financial field, based on the number of overdue premiums and credit score characteristics, high-risk insured people can be located.

[0103] The solution utilizes a dual-supervision model system to predict risk throughout the entire lifecycle. In the pre-complaint phase, the first supervisory model integrates features such as age, occupation, and historical complaint rates to predict the probability of customer complaints. For example, for individuals aged 60 and above living in rural areas and new health insurance users, the model predicts a 75% complaint risk, triggering preemptive intervention measures such as manual telephone follow-up and illustrated explanations of the terms and conditions, thereby reducing the incidence of such complaints. In the post-complaint phase, the second supervisory model analyzes metrics such as complaint amount, sentiment polarity, and past handling records, outputting three levels of difficulty: high, medium, and low. In the auto insurance scenario, complaints with a disputed amount greater than 150,000 yuan and containing keywords such as 'claim rejection' or 'unclear liability' are classified as high difficulty and automatically assigned to a team of experts, shortening the processing cycle and improving customer satisfaction.

[0104] Efficient dispute resolution is achieved through intelligent interception and tiered resolution strategies: For uninsured, high-risk customers, a dual interception approach of "random forest model + insurance knowledge test" is employed. For example, in the financial sector, customers with "less than one year of investment experience and unqualified risk assessments" are required to complete simulated trading training before purchasing insurance, reducing complaints due to misunderstandings of terms. For insured, high-risk customers, guidance is provided through "logistic regression model + surrender benefit calculation," improving the surrender conversion rate for health insurance complaints related to "coverage disputes." A dynamic matching system of "difficulty level - personnel skills - speech templates" is established. Low-difficulty complaints are handled by junior staff using standardized speech (such as "claim materials list text message"), improving efficiency. High-difficulty complaints trigger a mediated settlement model, integrating "disputed amount × industry benchmark ratio" to generate a recommended settlement fee. Through a three-level approval process (finance + legal + senior management) documented on the blockchain, the settlement fee payment cycle is shortened and ineffective expenditures are reduced.

[0105] Through the dynamic execution of intelligent interception and resolution strategies, a dual mechanism of "model prediction + rule verification" is employed to address uninsured high-risk customers. A logistic regression model is used to generate personalized cancellation plans for insured high-risk customers. For complex complaints, the settlement model integrates data such as the dispute amount and customer value to generate recommended settlement funds. Compliance is ensured through a three-tiered approval process backed by blockchain evidence. For example, in the auto insurance scenario, for complaints involving large and emotionally charged disputes, the entire process, from expert assignment to settlement fund payment, is shortened, significantly improving processing efficiency. In the healthcare sector, the generated settlement plan can be directly integrated with HL7 / FHIR-compliant medical information systems, providing standardized data support for insurance claims and customer service. In the fintech sector, Basel III-compliant decision logs can be seamlessly integrated into risk control systems, facilitating automated compliance reviews. Through multi-stage technology integration and industry-specific adaptation, this invention effectively addresses issues such as insufficient data utilization, reliance on experience in decision-making, and inefficient processes in customer complaint management, providing a highly efficient and intelligent complaint handling paradigm for related industries.

[0106] In summary, the solution implemented by the embodiment of the present invention can solve the problems in the prior art of being unable to predict complaint risks in advance and being unable to recommend resolution strategies based on the assessed resolution difficulty.

[0107] It should be understood that the order of execution of the steps in the above embodiments does not necessarily imply a specific order of execution. The order of execution of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention. The software tools or components not provided by our company that appear in the embodiments of this application are merely examples and do not represent actual use.

[0108] In one embodiment, a device for resolving insurance policy complaints is provided, and the device for resolving insurance policy complaints corresponds one-to-one with the method for resolving insurance policy complaints in the above embodiment. Figure 8 As shown, the insurance policy complaint resolution device includes an acquisition module 810, a first prediction module 820, an implementation module 830, a second prediction module 840, and a recommendation module 850. The functional modules are described in detail as follows:

[0109] An acquisition module 810 is configured to acquire customer data and determine a first status of the customer with respect to a target insurance policy based on the customer data, wherein the first status includes a non-complaint status and a complaint status;

[0110] A first prediction module 820 is configured to construct a first supervision model to predict the complaint risk of customers who have not filed a complaint, and obtain a complaint risk prediction result for the customers who have not filed a complaint;

[0111] An implementation module 830 is configured to implement an interception strategy based on the complaint risk prediction result;

[0112] The second prediction module 840 is used to construct a second supervision model to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtain a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint;

[0113] The recommendation module 850 is configured to recommend a resolution strategy based on the resolution difficulty level prediction result.

[0114] In one embodiment, the first prediction module 820 is specifically configured to:

[0115] Filtering the target policies for policies covered within a specified time period;

[0116] The target variable is defined as whether a complaint occurs within a preset time after the policy is underwritten;

[0117] Extract static attributes and dynamic behavior characteristics of pre-insurance customers. The static attributes include customer occupation and age, and the dynamic behavior characteristics include customer historical complaint rates and channel negative feedback data.

[0118] Divide the target policy into a training set and a validation set, and determine the model threshold based on the validation set performance;

[0119] Customers with a probability higher than the threshold are marked as high-risk customers.

[0120] In one embodiment, the implementation module 830 is specifically configured to:

[0121] determining a second status of the customer with respect to the target insurance policy based on the customer data, the second status including an uninsured state and an insured state;

[0122] A third supervision model is constructed to intercept the high-risk customers who are in the uninsured state from purchasing insurance, and to guide the high-risk customers who are in the insured state to cancel their insurance.

[0123] In one embodiment, the second prediction module 840 is specifically configured to:

[0124] Filter the customers who have filed complaints.

[0125] Define the target variable as whether there is another complaint within a preset time after the complaint;

[0126] Extracting attributes and historical behavior correlation features of the customer's complaints;

[0127] Train the solution difficulty prediction model and output the solution difficulty level as high, medium, and low difficulty labels.

[0128] In one embodiment, the recommendation module 850 is specifically configured to:

[0129] Allocate the customers who have complained to junior, intermediate and expert level resolution personnel according to the difficulty level of resolution;

[0130] Based on the characteristics of customers who have complained and historical resolution data, recommend the most successful strategy to the resolution personnel;

[0131] For customers with high-difficulty complaints, the settlement model is used to determine whether settlement funds are needed, and experts are assigned or a settlement fund plan is initiated based on the results.

[0132] In one embodiment, the recommendation module 850 is further configured to:

[0133] Analyze the success rate of historical sales pitches under different customer characteristics;

[0134] Match the optimal sales pitch based on the current customer's gender, age, and complaint channel characteristics;

[0135] Dynamically adjust the priority of recommended speech techniques based on the skill tags of the resolution personnel.

[0136] For customers with high-difficulty complaints, predict whether settlement funds are needed;

[0137] If it is predicted that no settlement is required, expert personnel will be assigned to handle the matter; if settlement is required, a recommended settlement fee will be generated and the payment process will be initiated.

[0138] In one embodiment, the recommendation module 850 is further configured to:

[0139] For customers with high-difficulty complaints, predict whether settlement funds are needed;

[0140] If it is predicted that no settlement is required, expert personnel will be assigned to handle the matter; if settlement is required, a recommended settlement fee will be generated and the payment process will be initiated.

[0141] The present invention provides a solution provided by a device for resolving insurance policy complaints. By constructing a first supervision model, high-complaint risk customers are identified in advance, the influx of high-risk customers is avoided, and the probability of complaints is reduced from the source. By constructing a second supervision model and analyzing the complaint attributes of customers who have already complained, the difficulty of resolution is graded, and resolution strategies are dynamically recommended to avoid ordinary cases occupying high-quality resources, shorten the processing cycle, reduce repeated customer complaints, and improve the success rate of resolution. A full-process digital model covering "risk prediction-interception-resolution" reduces reliance on manual experience and improves the objectivity and consistency of decision-making. It is not only applicable to traditional insurance business, but can also be extended to cross-border scenarios such as medical health, financial technology, etc. (such as health insurance, Internet insurance). Through cross-domain data fusion and model adaptation, the problem of complaint management in complex business scenarios can be effectively solved.

[0142] In summary, the solution implemented by the above-mentioned method for resolving insurance policy complaints can solve the problems that existing technologies cannot predict complaint risks in advance and cannot recommend resolution strategies based on the assessment of resolution difficulty.

[0143] The specific definition of the insurance policy complaint resolution device can be found in the definition of the insurance policy complaint resolution method above and will not be repeated here. Each module in the aforementioned insurance policy complaint resolution device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of these modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each of these modules.

[0144] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 9 As shown. The computer device includes a processor, a memory, a network interface and a database 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 and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it implements the functions or steps of the service-side of a method for recommending an optimal strategy.

[0145] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 10 As shown. The computer device includes a processor, memory, network interface, display screen, and 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 computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of the client side of a method for recommending an optimal strategy.

[0146] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed:

[0147] Acquiring customer data, and determining a first status of the customer with respect to a target insurance policy based on the customer data, the first status including a non-complaint status and a complained status;

[0148] Constructing a first supervision model to predict the complaint risk of customers who have not filed a complaint, and obtaining a complaint risk prediction result for the customers who have not filed a complaint;

[0149] Implement interception strategies based on the complaint risk prediction results;

[0150] Constructing a second supervision model to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtaining a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint;

[0151] A resolution strategy is recommended based on the resolution difficulty level prediction result.

[0152] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0153] Acquiring customer data, and determining a first status of the customer with respect to a target insurance policy based on the customer data, the first status including a non-complaint status and a complained status;

[0154] Constructing a first supervision model to predict the complaint risk of customers who have not filed a complaint, and obtaining a complaint risk prediction result for the customers who have not filed a complaint;

[0155] Implement interception strategies based on the complaint risk prediction results;

[0156] Constructing a second supervision model to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtaining a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint;

[0157] A resolution strategy is recommended based on the resolution difficulty level prediction result.

[0158] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can be found in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0159] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0160] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0161] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for resolving insurance policy complaints, characterized by: include: Acquiring customer data, and determining a first status of the customer with respect to a target insurance policy based on the customer data, the first status including a non-complaint status and a complained status; Constructing a first supervision model to predict the complaint risk of customers who have not filed a complaint, and obtaining a complaint risk prediction result for the customers who have not filed a complaint; Implement interception strategies based on the complaint risk prediction results; Constructing a second supervision model to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtaining a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint; A resolution strategy is recommended based on the resolution difficulty level prediction result.

2. The method for resolving insurance policy complaints according to claim 1, characterized in that: The first supervision model is constructed to predict the complaint risk of the customers who have not filed a complaint, and obtain the complaint risk prediction results of the customers who have not filed a complaint; including: Filter target policies that are insured within a specified time period; The target variable is defined as whether a complaint occurs within a preset time after the policy is underwritten; Extract static attributes and dynamic behavior characteristics of pre-insurance customers. The static attributes include customer occupation and age, and the dynamic behavior characteristics include customer historical complaint rates and channel negative feedback data. Divide the target policy into a training set and a validation set, and determine the model threshold based on the validation set performance; Customers with a probability higher than the threshold are marked as high-risk customers.

3. The method for resolving insurance policy complaints according to claim 2, characterized in that: The implementation of interception strategies based on the complaint risk prediction results includes: determining a second status of the customer with respect to the target insurance policy based on the customer data, the second status including an uninsured state and an insured state; A third supervision model is constructed to intercept the high-risk customers who are in the uninsured state from purchasing insurance, and to guide the high-risk customers who are in the insured state to cancel their insurance.

4. The method for resolving insurance policy complaints according to claim 1, characterized in that: The second supervision model is constructed to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, thereby obtaining a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint; the prediction result includes: Filter the customers who have filed complaints. Define the target variable as whether there is another complaint within a preset time after the complaint; Extracting attributes and historical behavior correlation features of the customer's complaints; Train the solution difficulty prediction model and output the solution difficulty level as high, medium, and low difficulty labels.

5. The method for resolving insurance policy complaints according to claim 4, characterized in that: The method of recommending a resolution strategy based on the resolution difficulty level prediction result includes: Allocate the customers who have complained to junior, intermediate and expert level resolution personnel according to the difficulty level of resolution; Based on the characteristics of customers who have complained and historical resolution data, recommend the most successful strategy to the resolution personnel; For customers with high-difficulty complaints, the settlement model is used to determine whether settlement funds are needed, and experts are assigned or a settlement fund plan is initiated based on the results.

6. The method for resolving insurance policy complaints according to claim 5, characterized in that: Based on the characteristics of the customers who have complained and the historical resolution data, recommend the highest success rate speech strategy to the resolution personnel; including: Analyze the success rate of historical sales pitches under different customer characteristics; Match the optimal sales pitch based on the current customer's gender, age, and complaint channel characteristics; Dynamically adjust the priority of recommended speech techniques based on the skill tags of the resolution personnel. For customers with high-difficulty complaints, predict whether settlement funds are needed; If it is predicted that no settlement is required, expert personnel will be assigned to handle the matter; if settlement is required, a recommended settlement fee will be generated and the payment process will be initiated.

7. The method for resolving insurance policy complaints according to claim 5, characterized in that: For clients with high-difficulty complaints, the assisted settlement model is used to determine whether assisted settlement funds are needed, and experts are assigned or assisted settlement fund plans are initiated based on the results; include: For customers with high-difficulty complaints, predict whether settlement funds are needed; If it is predicted that no settlement is required, expert personnel will be assigned to handle the matter; if settlement is required, a recommended settlement fee will be generated and the payment process will be initiated.

8. A device for resolving insurance policy complaints, characterized in that: include: an acquisition module, configured to acquire customer data and determine a first status of the customer with respect to a target insurance policy based on the customer data, wherein the first status includes a non-complaint status and a complained status; A first prediction module is used to construct a first supervision model to predict the complaint risk of customers in the non-complaint state, and obtain a complaint risk prediction result of the customers in the non-complaint state; An implementation module, used to implement an interception strategy based on the complaint risk prediction results; A second prediction module is used to construct a second supervision model to predict the difficulty level of resolving the complaint for the customer who has filed a complaint, and obtain a prediction result of the difficulty level of resolving the complaint for the customer who has filed a complaint; The recommendation module is used to recommend a resolution strategy based on the prediction result of the resolution difficulty level.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method for resolving insurance policy complaints as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for resolving insurance policy complaints as described in any one of claims 1 to 7 are implemented.

Images