Customer complaint risk identification and analysis method based on large model and quality inspection

Abstract

The invention relates to a customer complaint risk identification and analysis method based on a large model and quality inspection. The method comprises the following steps: a, collecting customer complaint verbal skills; b, configuring a quality inspection rule of the quality inspection system and a calculation formula of customer complaint; c, the customer service call center accesses an ASR engine to convert the voice flow into text content; d, inputting the text content into the large model and a quality inspection system; and e, according to the score returned by the large model and the score returned by the quality inspection system, different verbal skills and coping strategies are returned by using the large model, the problem of the customer is solved, customer complaints are reduced, and different customer complaint level labels are added to the customer. The customer complaint risk recognition and analysis method based on the large model and quality inspection has the advantages that efficient recognition, analysis and processing of customer complaints can be achieved through advanced technical means such as NLP, deep learning and automatic speech recognition (ASR), and the customer complaints can be effectively recognized, analyzed and processed through automatic speech-text conversion and text analysis. And the time and the cost of manual customer complaint handling are greatly reduced.

Description

Technical Field

[0001] The present invention relates to a customer complaint risk identification and analysis method based on a large model and quality inspection, and belongs to the field of financial derivative customer complaint risks. Background Art

[0002] In the financial customer complaint scenario, early identification and analysis of customer complaints is one of the important risk control methods. However, there are some problems with traditional customer complaint risk identification, such as how to identify risks in advance and adjust risk strategies in a timely manner. The disadvantages of existing technologies are mainly reflected in the following aspects:

[0003] Technical response delay: The traditional customer complaint handling process is often unable to achieve real-time monitoring and rapid response to customer complaint information, resulting in delays in risk identification and handling.

[0004] Insufficient data integration: Customer complaint information is scattered across different platforms and records, and there is a lack of effective data integration tools, which limits the comprehensive analysis and application of data and makes it difficult to fully identify and manage customer complaint risks.

[0005] Technical risks and security issues: The rapid development of financial technology has brought new security challenges, and existing security technologies have failed to keep up, resulting in the reduction of financial technology security performance and the expansion of the scope of security issues.

[0006] Regulatory uncertainty: Non-bank financial products / services, decentralized finance, and digital assets are penetrating financial services in an environment of near-zero regulation, posing unique challenges to financial stability, secure transactions, and general customer protection.

[0007] Customer data privacy controls are lagging and inconsistent: With the increase in financial data and the diversification of transaction patterns, data security and privacy protection issues have become increasingly prominent. Existing control measures are lagging and inconsistent, increasing the risk of customer data leakage and abuse.

[0008] Insufficient identification at the personnel level: Lack of establishing a complaint prevention mechanism, intervening in the customer complaint process in advance to reduce the occurrence of customer complaints, and providing professional skills training to employees to improve their ability to handle customer complaints.

[0009] Limited data analysis and prediction capabilities: Although CRM systems and data mining technologies, such as cluster analysis models, are used to analyze customer data and identify complaint risk points, the ability to establish customer churn tendency warning models, determine customer churn warning points, and formulate recovery strategies is limited.

[0010] These drawbacks show that although the development of financial technology has brought new possibilities for identifying customer complaint risks, it has also brought new challenges and risks, requiring financial institutions to take more effective technical and management measures to deal with them. Summary of the invention

[0011] In order to overcome the defects of the prior art, the present invention provides a customer complaint risk identification and analysis method based on a large model and quality inspection. The technical solution of the present invention is:

[0012] A method for identifying and analyzing customer complaint risks based on a big model and quality inspection, comprising the following steps: a. collecting customer complaint scripts, importing the scripts into a big model for training, and configuring a customer complaint identification calculation formula;

[0013] b. Configure the quality inspection rules of the quality inspection system and the calculation formula for customer complaints;

[0014] c. The customer service call center accesses the ASR engine to convert voice streams into text content;

[0015] d. Input the text content into the big model and the quality inspection system, configure prompt words in the big model, and calculate the corresponding score according to the configured calculation formula and return it;

[0016] e. Calculate the average value based on the scores returned by the big model and the scores returned by the quality inspection system. The calculated score is obtained through the conversation between the customer and the customer service. The big model is used to return different words and response strategies to solve customer problems, reduce customer complaints, and add different customer complaint level labels to customers.

[0017] The step a is specifically:

[0018] 1.1 Collection of customer complaint texts: Collect text data of customer complaints and feedback through customer service call centers, online chats, and social media channels;

[0019] The collected customer complaint scripts include but are not limited to customers’ dissatisfaction, complaints, and suggestions about products or services, so that the collected script data covers issues and terminology in the financial industry;

[0020] 1.2 Data preprocessing: Clean the collected dialogues, including removing irrelevant information, formatting text, and correcting typos; annotate the dialogues, mark key information, and classify the dialogues;

[0021] 1.3 Importing large models for training: Importing pre-processed speech data into the large model for training; training the model to identify key information in customer complaint speech, and using deep learning technology to improve the model's understanding and generation capabilities;

[0022] 1.4 Configure the calculation formula for customer complaint identification: According to business needs and historical data analysis, determine the key factors for customer complaint risk identification; assign weights to each key factor. The weight allocation is based on historical data, customer behavior analysis and psychological principles. The calculation formula is:

[0023] Customer complaint risk score = financial regulatory nouns × w1 + dirty words × w2 + emotion recognition × w3 + decibel × w4 + accompanying keywords × w5 Customer complaint risk score = financial regulatory nouns × w1 + dirty words × w2 + emotion recognition × w3 + decibel × w4 + accompanying keywords × w5;

[0024] Among them, w1,w2,w3,w4,w5w1,w2,w3,w4,w5 are the weight scores of each factor respectively;

[0025] 1.5 Model verification and optimization: Use the verification set to test the trained model to evaluate the accuracy and reliability of the model; adjust the model parameters and weights according to the verification results to optimize the model performance.

[0026] 1.6 Model deployment: Deploy the trained model to the customer service system for real-time identification and analysis of customer complaints; update the model regularly to adapt to new customer complaints and market changes.

[0027] The step b is specifically:

[0028] 2.1 Define quality inspection goals: determine key performance indicators (KPIs), including first-time resolution rate, average processing time, and customer satisfaction score;

[0029] 2.2 Design quality inspection rules:

[0030] Compliance rules: Ensure customer service representatives comply with company policies and industry regulations;

[0031] Service quality rules: including response time, problem solving efficiency, and the positivity of customer feedback; Interaction quality rules: evaluate the communication skills of customer service representatives;

[0032] 2.3. Formulate the customer complaint calculation formula:

[0033] Factor identification: Identify factors that influence the severity of customer complaints;

[0034] Weighting: Assign weights to each factor based on historical data analysis, business impact, and customer feedback;

[0035] Formula construction: Construct a comprehensive formula: Customer complaint risk score = (emotion score × w1) + (problem type score × w2) + (scope of impact score × w3) Customer complaint risk score = (emotion score × w1) + (problem type score × w2) + (scope of impact score × w3) Where w1, w2, w3w1, w2, w3 are weights assigned according to the importance of factors;

[0036] The step c is specifically as follows:

[0037] 3.1 The customer service call center achieves speech-to-text conversion by integrating an automatic speech recognition (ASR) engine. The ASR engine processes real-time speech streams and converts them into written text for further analysis and processing;

[0038] 3.2 Select the ASR engine based on business needs and resources;

[0039] 3.3 Before speech-to-text conversion, the audio signal is pre-processed, including noise reduction and format conversion, to meet the requirements of the ASR system;

[0040] 3.4 The ASR engine extracts key acoustic features from the audio to provide basic data for recognition;

[0041] 3.5 Acoustic model and language model: The acoustic model is used to identify the correspondence between acoustic features and speech units, and the language model is combined to predict and correct word sequences to improve recognition accuracy; 3.6 The output of the acoustic and language models is converted into a text sequence through the decoder to complete the conversion of speech to text;

[0042] 3.7 Real-time transcription and post-processing: The ASR engine transcribes the conversation in real time, receiving the voice and returning the text.

[0043] The step d is specifically as follows:

[0044] 4.1 Ensure that the text content converted from the ASR engine is in the correct and clear format, so that it can be easily processed by the large model and quality inspection system;

[0045] 4.2 Input the preprocessed text content into the large model to understand the context and semantics of the text;

[0046] 4.3 Input the text content into the quality inspection system, which is used to evaluate the service quality and customer complaint handling effect of the customer service;

[0047] 4.3 Apply the configured calculation formula to score the text content;

[0048] 4.4 Score calculation: The big model and quality inspection system calculate the corresponding score based on the input text content and the configured calculation formula;

[0049] 4.5 Result return: The calculated score is returned to the customer service system so that the customer service representative can take appropriate actions based on the score.

[0050] The step e is specifically as follows:

[0051] 5.1 Receive the scores returned by the big model and the quality inspection system, compare the scores of the two systems, identify the differences, and conduct preliminary analysis to determine the causes;

[0052] 5.2 Calculate the average of the two scores to obtain a comprehensive customer complaint risk score; specifically:

[0053] Comprehensive risk score = (large model score + quality inspection system score) ÷ 2;

[0054] 5.3 Use the comprehensive risk score to analyze the conversation between the customer and the customer service staff to identify key problem points and the customer's emotional state;

[0055] 5.4 Based on the results of the conversation analysis, the big model generates or recommends corresponding words and response strategies, which are designed to solve specific problems of customers, alleviate negative emotions, and improve customer satisfaction;

[0056] 5.5 Customer complaint handling: Customer service representatives interact with customers and solve problems based on the words and strategies provided by the big model; monitor the progress of the conversation and adjust the strategy when necessary to ensure that the problem is effectively resolved;

[0057] 5.6 Customer complaint level label assignment: Different customer complaint level labels are assigned to customer issues based on the comprehensive risk score and conversation results;

[0058] 5.7 Feedback and Optimization: Collect customer feedback, evaluate the effectiveness of sales pitches and strategies, and continuously optimize the big model and quality inspection system based on the feedback results to improve the accuracy and efficiency of customer complaint handling.

[0059] The advantages of the present invention are:

[0060] The core advantage of the customer complaint risk identification and analysis method based on large models and quality inspection is that it can achieve efficient identification, analysis and processing of customer complaints through advanced technical means, such as natural language processing (NLP), deep learning, automatic speech recognition (ASR), etc. It has the following advantages:

[0061] Automation and efficiency improvement: Through automated speech-to-text and text analysis, the time and cost of manually handling customer complaints are greatly reduced.

[0062] Accurate identification: Using big models to deeply learn and understand customer complaint language can more accurately identify customer emotions, intentions, and problem types.

[0063] Real-time feedback: The system can analyze customer service conversations in real time, provide timely feedback and response strategies, and help resolve issues quickly.

[0064] Risk assessment: By integrating the scores of the big model and the quality inspection system, the customer complaint risk can be quantitatively assessed to help companies better manage and respond.

[0065] Personalized service: Generate personalized words and response strategies based on the customer's specific problems and emotional state to improve customer satisfaction.

[0066] Data-driven decision-making: By collecting and analyzing large amounts of customer complaint data, companies can make more scientific decisions based on the data.

[0067] Continuous optimization: The system can continuously learn and optimize based on feedback to improve the accuracy and efficiency of handling customer complaints. DETAILED DESCRIPTION

[0068] The present invention will be further described below in conjunction with specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are exemplary only and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solution of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, but these modifications and replacements all fall within the scope of protection of the present invention.

[0069] The present invention relates to a customer complaint risk identification and analysis method based on a large model and quality inspection, comprising the following steps:

[0070] a. Collect customer complaint scripts, import them into the big model for training, and configure the customer complaint recognition calculation formula;

[0071] b. Configure the quality inspection rules of the quality inspection system and the calculation formula for customer complaints;

[0072] c. The customer service call center accesses the ASR engine to convert voice streams into text content;

[0073] d. Input the text content into the big model and the quality inspection system, configure prompt words in the big model, and calculate the corresponding score according to the configured calculation formula and return it;

[0074] e. Calculate the average value based on the scores returned by the big model and the scores returned by the quality inspection system. The calculated score is obtained through the conversation between the customer and the customer service. The big model is used to return different words and response strategies to solve customer problems, reduce customer complaints, and add different customer complaint level labels to customers.

[0075] The step a is specifically:

[0076] 1.1 Collection of customer complaint texts: Collect text data of customer complaints and feedback through customer service call centers, online chats, and social media channels;

[0077] The collected customer complaint scripts include but are not limited to customers’ dissatisfaction, complaints, and suggestions about products or services, so that the collected script data covers issues and terminology in the financial industry;

[0078] 1.2 Data preprocessing: Clean the collected words, including removing irrelevant information, formatting text, and correcting typos;

[0079] Annotate the conversations and mark out key information, such as customer sentiment, question type, keywords, etc.

[0080] Classify the dialogue techniques, such as by the severity of the problem, the type of problem, etc.;

[0081] 1.3 Importing large models for training: Importing preprocessed speech data into large models, such as natural language processing (NLP) models, for training;

[0082] Train the model to identify key information in customer complaint dialogue, such as sentiment, intent, question type, etc. Use deep learning techniques, such as recurrent neural networks (RNN), long short-term memory networks (LSTM), or Transformer models to improve the model's understanding and generation capabilities;

[0083] 1.4 Configure the customer complaint identification calculation formula: According to business needs and historical data analysis, determine the key factors for customer complaint risk identification, such as financial regulatory terms, dirty words, emotion recognition, decibels, and accompanying keywords;

[0084] Assign weights to each key factor based on historical data, customer behavior analysis, and psychological principles. The calculation formula is:

[0085] Customer complaint risk score = financial regulatory nouns × w1 + dirty words × w2 + emotion recognition × w3 + decibel × w4 + accompanying keywords × w5 Customer complaint risk score = financial regulatory nouns × w1 + dirty words × w2 + emotion recognition × w3 + decibel × w4 + accompanying keywords × w5;

[0086] Among them, w1,w2,w3,w4,w5w1,w2,w3,w4,w5 are the weight scores of each factor respectively;

[0087] 1.5 Model verification and optimization: Use the verification set to test the trained model to evaluate the accuracy and reliability of the model; adjust the model parameters and weights according to the verification results to optimize the model performance.

[0088] 1.6 Model deployment: Deploy the trained model to the customer service system for real-time identification and analysis of customer complaints; update the model regularly to adapt to new customer complaints and market changes.

[0089] The step b is specifically:

[0090] 2.1 Define quality inspection goals: determine key performance indicators (KPIs), including first-time resolution rate, average processing time, and customer satisfaction score;

[0091] 2.2 Design quality inspection rules:

[0092] Compliance rules: Ensure that customer service representatives comply with company policies and industry regulations; such as protecting customer privacy and not using inappropriate language.

[0093] Service quality rules: including response time, problem solving efficiency, and the degree of positivity of customer feedback; Interaction quality rules: evaluating the customer service representative’s communication skills, such as empathy, patience, and the transfer of professional knowledge;

[0094] 2.3. Formulate the customer complaint calculation formula:

[0095] Factor identification: Identify factors that affect the severity of customer complaints, such as customer sentiment, problem type, scope of impact, etc.

[0096] Weighting: Assign weights to each factor based on historical data analysis, business impact, and customer feedback;

[0097] Formula construction: Construct a comprehensive formula: Customer complaint risk score = (emotion score × w1) + (problem type score × w2) + (scope of impact score × w3) Customer complaint risk score = (emotion score × w1) + (problem type score × w2) + (scope of impact score × w3) Where w1, w2, w3w1, w2, w3 are weights assigned according to the importance of factors;

[0098] The step c is specifically as follows:

[0099] 3.1 The customer service call center achieves speech-to-text conversion by integrating an automatic speech recognition (ASR) engine. The ASR engine processes real-time speech streams and converts them into written text for further analysis and processing;

[0100] 3.2 Select an ASR engine based on business needs and resources. For example, you can choose open source ASR engines such as Julius, ESPnet, Tensorflow ASR, etc. These engines support multiple languages ​​and allow custom models to improve recognition accuracy. When deploying an ASR engine, consider using multi-cluster, multi-IDC, and multi-vendor ASR products to improve availability and fault tolerance.

[0101] 3.3 Before speech-to-text conversion, the audio signal is pre-processed, including noise reduction and format conversion, to meet the requirements of the ASR system;

[0102] 3.4 The ASR engine extracts key acoustic features from the audio, such as MFCC (Mel-frequency cepstral coefficients), providing basic data for recognition;

[0103] 3.5 Acoustic model and language model: The acoustic model is used to identify the correspondence between acoustic features and speech units, and the language model is combined to predict and correct word sequences to improve recognition accuracy; 3.6 The output of the acoustic and language models is converted into a text sequence through the decoder to complete the conversion of speech to text;

[0104] 3.7 Real-time transcription and post-processing: The ASR engine transcribes the conversation in real time, receiving the voice and returning the text. The recognition effect is optimized and the personalized accuracy is improved to adapt to the jargon, language, accent and dialect of specific industries.

[0105] The step d is specifically as follows:

[0106] 4.1 Ensure that the text content converted from the ASR engine is in the correct and clear format to facilitate processing by large models and quality inspection systems; perform necessary preprocessing on the text, such as word segmentation, removal of stop words, part-of-speech tagging, etc., to improve the model's understanding and processing efficiency;

[0107] 4.2 Input the preprocessed text content into the big model, which is a pre-trained language model such as BERT, GPT, etc., which is used to understand the context and semantics of the text.

[0108] Configure prompts in large models. These prompts are carefully designed to guide the model to focus on key information in the text, such as customer sentiment, question type, etc.

[0109] 4.3 Input the text content into the quality inspection system, which is used to evaluate the customer service quality and customer complaint handling effect; the quality inspection system analyzes the text content according to preset rules and models to identify key issues and service quality indicators.

[0110] 4.3 Apply the configured calculation formula to score the text content;

[0111] 4.4 Score calculation: The big model and quality inspection system calculate the corresponding score based on the input text content and the configured calculation formula; the score reflects the severity, urgency and possible impact of the customer complaint.

[0112] 4.5 Result return: The calculated score is returned to the customer service system so that the customer service representative can take appropriate actions based on the score. The score can also be used to trigger automated processes, such as escalating serious issues, notifying management, etc.

[0113] The step e is specifically as follows:

[0114] 5.1 Receive the scores returned by the big model and the quality inspection system, compare the scores of the two systems, identify the differences, and conduct preliminary analysis to determine the causes;

[0115] 5.2 Calculate the average of the two scores to obtain a comprehensive customer complaint risk score; specifically:

[0116] Comprehensive risk score = (large model score + quality inspection system score) ÷ 2;

[0117] 5.3 Use the comprehensive risk score to analyze the conversation between the customer and the customer service representative to identify key problem points and the customer’s emotional state.

[0118] Apply natural language processing techniques, such as sentiment analysis and intent recognition, to gain deep insights into the nature of conversations.

[0119] 5.4 Based on the results of the conversation analysis, the big model generates or recommends corresponding words and response strategies, which are designed to solve specific problems of customers, alleviate negative emotions, and improve customer satisfaction;

[0120] 5.5 Customer complaint handling: Customer service representatives interact with customers and solve problems based on the words and strategies provided by the big model; monitor the progress of the conversation and adjust the strategy when necessary to ensure that the problem is effectively resolved;

[0121] 5.6 Customer complaint level label assignment: Based on the comprehensive risk score and the results of the conversation, different customer complaint level labels are assigned to the customer's problem, such as "low risk", "medium risk", "high risk", etc. These labels help prioritize more serious issues and provide data support for future customer complaint prevention and service quality improvement.

[0122] 5.7 Feedback and Optimization: Collect customer feedback, evaluate the effectiveness of sales pitches and strategies, and continuously optimize the big model and quality inspection system based on the feedback results to improve the accuracy and efficiency of customer complaint handling.

[0123] The working principle of the present invention is as follows:

[0124] Data collection and preprocessing (step a): Collect customer complaints from multiple channels, clean, label and classify them, and then import them into the big model for training.

[0125] Model training and configuration (step a): Use NLP model training to identify key information in customer complaint speech and configure the customer complaint identification calculation formula.

[0126] Quality inspection system configuration (step b): define quality inspection objectives, design quality inspection rules, and develop a customer complaint calculation formula.

[0127] Speech-to-text (step c): The speech stream is converted into text content through the ASR engine, preprocessed, acoustic features are extracted, and recognized using acoustic models and language models.

[0128] Text analysis and scoring (step d): The preprocessed text is input into the big model and quality inspection system, and the calculation formula is applied to calculate the score of the customer complaint.

[0129] Comprehensive scoring and response strategies (step e): Receive the scores of the two systems, calculate the average, analyze the conversation content, and generate or recommend speech and response strategies.

[0130] Customer complaint handling and label assignment (step e): Customer service representatives interact with customers based on recommended scripts and strategies, solve problems, and assign customer complaint level labels based on the results.

[0131] Feedback and optimization (step e): Collect customer feedback, evaluate the effectiveness of sales pitches and strategies, and optimize models and systems based on feedback results.

[0132] Through this method, companies can not only improve the efficiency and accuracy of customer complaint handling, but also continuously improve service quality and customer satisfaction through data analysis and model optimization.

[0133] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A customer complaint risk identification and analysis method based on a large model and quality inspection, characterized in that: The following steps are involved: a. Collect customer complaint scripts, import them into the big model for training, and configure the customer complaint recognition calculation formula; b. Configure the quality inspection rules of the quality inspection system and the calculation formula for customer complaints; c. The customer service call center accesses the ASR engine to convert voice streams into text content; d. Input the text content into the big model and the quality inspection system, configure prompt words in the big model, and calculate the corresponding score according to the configured calculation formula and return it; e. Calculate the average value based on the scores returned by the big model and the scores returned by the quality inspection system. The calculated score is obtained through the conversation between the customer and the customer service. The big model is used to return different words and response strategies to solve customer problems, reduce customer complaints, and add different customer complaint level labels to customers.

2. According to claim 1, a method for identifying and analyzing customer complaint risks based on a large model and quality inspection is characterized in that: The step a is specifically: 1.1 Collection of customer complaint texts: Collect text data of customer complaints and feedback through customer service call centers, online chats, and social media channels; The collected customer complaint scripts include but are not limited to customers’ dissatisfaction, complaints, and suggestions about products or services, so that the collected script data covers issues and terminology in the financial industry; 1.2 Data preprocessing: Clean the collected words, including removing irrelevant information, formatting text, and correcting typos; Annotate the dialogues, mark out key information, and classify the dialogues; 1.3 Importing the big model for training: Import the pre-processed speech data into the big model for training; Train the model to identify key information in customer complaint dialogues and use deep learning technology to improve the model's understanding and generation capabilities; 1.4 Configure the calculation formula for customer complaint identification: According to business needs and historical data analysis, determine the key factors for customer complaint risk identification; assign weights to each key factor. The weight allocation is based on historical data, customer behavior analysis and psychological principles. The calculation formula is: Customer complaint risk score = financial regulatory nouns × w1 + dirty words × w2 + emotion recognition × w3 + decibel × w4 + accompanying keywords × w5 Customer complaint risk score = financial regulatory nouns × w1 + dirty words × w2 + emotion recognition × w3 + decibel × w4 + accompanying keywords × w5; Among them, w1,w2,w3,w4,w5w1,w2,w3,w4,w5 are the weight scores of each factor respectively; 1.5 Model verification and optimization: Use the verification set to test the trained model to evaluate the accuracy and reliability of the model; adjust the model parameters and weights according to the verification results to optimize the model performance. 1.6 Model deployment: Deploy the trained model to the customer service system for real-time identification and analysis of customer complaints; update the model regularly to adapt to new customer complaints and market changes.

3. According to claim 1, a method for identifying and analyzing customer complaint risks based on a large model and quality inspection is characterized in that: The step b is specifically: 2.1 Define quality inspection goals: determine key performance indicators (KPIs), including first-time resolution rate, average processing time, and customer satisfaction score; 2.2 Design quality inspection rules: Compliance rules: Ensure customer service representatives comply with company policies and industry regulations; Service quality rules: including response time, problem solving efficiency, and the positivity of customer feedback; Interaction quality rules: Evaluate the customer service representative’s communication skills; 2.

3. Formulate the customer complaint calculation formula: Factor identification: Identify factors that influence the severity of customer complaints; Weighting: Assign weights to each factor based on historical data analysis, business impact, and customer feedback; Formula construction: Construct a comprehensive formula: Customer complaint risk score = (emotion score × w1) + (problem type score × w2) + (impact range score × w3) Customer complaint risk score = (emotion score × w1) + (problem type score × w2) + (impact range score × w3) Where w1, w2, w3w1, w2, w3 are weights assigned according to the importance of factors.

4. According to claim 1, a method for identifying and analyzing customer complaint risks based on a large model and quality inspection is characterized in that: The step c is specifically as follows: 3.1 The customer service call center achieves speech-to-text conversion by integrating an automatic speech recognition (ASR) engine. The ASR engine processes real-time speech streams and converts them into written text for further analysis and processing; 3.2 Select the ASR engine based on business needs and resources; 3.3 Before speech-to-text conversion, the audio signal is pre-processed, including noise reduction and format conversion, to meet the requirements of the ASR system; 3.4 The ASR engine extracts key acoustic features from the audio to provide basic data for recognition; 3.5 Acoustic model and language model: The acoustic model is used to identify the correspondence between acoustic features and speech units, and the language model is combined to predict and correct word sequences to improve recognition accuracy; 3.6 The output of the acoustic and language models is converted into a text sequence through the decoder to complete the conversion of speech to text; 3.7 Real-time transcription and post-processing: The ASR engine transcribes the conversation in real time, receiving the voice and returning the text.

5. The method for identifying and analyzing customer complaint risks based on a large model and quality inspection according to claim 1 is characterized in that: The step d is specifically as follows: 4.1 Ensure that the text content converted from the ASR engine is in the correct and clear format, so that it can be easily processed by the large model and quality inspection system; 4.2 Input the preprocessed text content into the large model to understand the context and semantics of the text; 4.3 Input the text content into the quality inspection system, which is used to evaluate the service quality and customer complaint handling effect of the customer service; 4.3 Apply the configured calculation formula to score the text content; 4.4 Score calculation: The big model and quality inspection system calculate the corresponding score based on the input text content and the configured calculation formula; 4.5 Result return: The calculated score is returned to the customer service system so that the customer service representative can take appropriate actions based on the score.

6. The method for identifying and analyzing customer complaint risks based on a large model and quality inspection according to claim 1 is characterized in that: The step e is specifically as follows: 5.1 Receive the scores returned by the big model and the quality inspection system, compare the scores of the two systems, identify the differences, and conduct preliminary analysis to determine the causes; 5.2 Calculate the average of the two scores to obtain a comprehensive customer complaint risk score; specifically: Comprehensive risk score = (large model score + quality inspection system score) ÷ 2; 5.3 Use the comprehensive risk score to analyze the conversation between the customer and the customer service staff to identify key problem points and the customer's emotional state; 5.4 Based on the results of the conversation analysis, the big model generates or recommends corresponding words and response strategies, which are designed to solve specific problems of customers, alleviate negative emotions, and improve customer satisfaction; 5.5 Customer complaint handling: Customer service representatives interact with customers and solve problems based on the words and strategies provided by the big model; monitor the progress of the conversation and adjust the strategy when necessary to ensure that the problem is effectively resolved; 5.6 Customer complaint level label assignment: Different customer complaint level labels are assigned to customer issues based on the comprehensive risk score and conversation results; 5.7 Feedback and Optimization: Collect customer feedback, evaluate the effectiveness of sales pitches and strategies, and continuously optimize the big model and quality inspection system based on the feedback results to improve the accuracy and efficiency of customer complaint handling.