Customer service dialogue quality inspection method and device, computer device and storage medium

Abstract

This application belongs to the field of artificial intelligence and relates to a customer service dialogue quality inspection method, apparatus, computer equipment, and storage medium. The method includes: acquiring dialogue text between a customer and a customer service representative; acquiring a training inference chain for the dialogue text, the training inference chain including at least two levels of quality inspection prompts and corresponding quality inspection labels for each prompt, the quality inspection prompts corresponding to quality inspection points in the dialogue text, and the quality inspection prompts at each level forming a progressive inference relationship; progressively training an initial dialogue quality inspection model based on the dialogue text, the quality inspection prompts at each level in the training inference chain, and their corresponding quality inspection labels to obtain a dialogue quality inspection model, the quality inspection prompts being used to guide the initial dialogue quality inspection model to generate prediction results; acquiring the dialogue text to be inspected between a target customer and a target customer service representative, and inputting the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service representative. This application improves the accuracy of customer service dialogue quality inspection.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a customer service dialogue quality inspection method, apparatus, computer equipment, and storage medium. Background Technology

[0002] In service industries such as finance, insurance, and banking, monitoring and evaluating customer service quality is crucial for improving customer satisfaction and optimizing service processes. Traditional quality control methods primarily involve manual review, evaluating customer service conversations based on pre-set standards. However, manual quality control is inefficient and struggles to process conversation data in real time. Furthermore, manual evaluation is highly subjective; different quality control personnel may provide different assessments of the same conversation, affecting the consistency and accuracy of the quality control process.

[0003] Although some automated dialogue quality inspection systems have emerged, they mainly rely on simple keyword matching and basic sentiment analysis, which has limitations in handling complex dialogues and understanding contextual information, resulting in inaccurate quality inspection results. Summary of the Invention

[0004] The purpose of this application is to provide a customer service dialogue quality inspection method, apparatus, computer equipment, and storage medium to solve the problem of low accuracy in customer service dialogue quality inspection.

[0005] To address the aforementioned technical problems, this application provides a customer service dialogue quality inspection method, employing the following technical solution:

[0006] Obtain the text of the conversation between the customer and customer service;

[0007] Obtain the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship.

[0008] Based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain a dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results.

[0009] Obtain the dialogue text between the target customer and the target customer service representative, and input the dialogue text into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service representative.

[0010] To address the aforementioned technical problems, this application also provides a customer service dialogue quality inspection device, which employs the following technical solution:

[0011] The text acquisition module is used to acquire the dialogue text between the customer and customer service.

[0012] The training acquisition module is used to acquire the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship.

[0013] The model training module is used to progressively train the initial dialogue quality inspection model based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels to obtain a dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results.

[0014] The dialogue quality inspection module is used to obtain the dialogue text to be inspected between the target customer and the target customer service representative, and input the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service representative.

[0015] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:

[0016] Obtain the text of the conversation between the customer and customer service;

[0017] Obtain the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship.

[0018] Based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain a dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results.

[0019] Obtain the dialogue text between the target customer and the target customer service representative, and input the dialogue text into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service representative.

[0020] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:

[0021] Obtain the text of the conversation between the customer and customer service;

[0022] Obtain the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship.

[0023] Based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain a dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results.

[0024] Obtain the dialogue text between the target customer and the target customer service representative, and input the dialogue text into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service representative.

[0025] Compared with existing technologies, the embodiments of this application have the following main advantages: They acquire the dialogue text between the customer and customer service representatives, as well as the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and corresponding quality inspection labels. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and each level of quality inspection prompt is a progressive inference relationship. Based on the dialogue text, the quality inspection prompts at each level in the training inference chain, and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain the dialogue quality inspection model. The inference chain enables the model to analyze the dialogue content in depth layer by layer. The progressive training method allows the model to not only evaluate the surface-level dialogue quality but also to deeply understand the logic, emotional expression, and service quality of the dialogue. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results, prompting the model to generate more accurate and comprehensive quality inspection results, covering everything from basic information checks to complex situational understanding. Acquiring the dialogue text to be inspected between the target customer and the target customer service representative and inputting it into the dialogue quality inspection model can automatically achieve customer service dialogue quality inspection and improve the accuracy of customer service dialogue quality inspection. Attached Figure Description

[0026] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0028] Figure 2 This is a flowchart of an embodiment of the customer service dialogue quality inspection method according to this application;

[0029] Figure 3 This is a schematic diagram of a structure of an embodiment of the customer service dialogue quality inspection device according to this application;

[0030] Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0032] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0033] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0034] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0035] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0036] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc.

[0037] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0038] It should be noted that the customer service dialogue quality inspection method provided in this application embodiment is generally executed by the server, and correspondingly, the customer service dialogue quality inspection device is generally set in the server.

[0039] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0040] Continue to refer to Figure 2 A flowchart of an embodiment of the customer service dialogue quality inspection method according to this application is shown. The customer service dialogue quality inspection method includes the following steps:

[0041] Step S201: Obtain the dialogue text between the customer and customer service.

[0042] In this embodiment, the customer service dialogue quality inspection method operates on an electronic device (e.g., Figure 1 The server shown can communicate with the terminal device via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, Wi-Fi connections, Bluetooth connections, Wi-Fi connections, Zigbee connections, UWB (ultra-Width band) connections, and other currently known or future wireless connection methods.

[0043] Specifically, during the model training phase, the dialogue text between the customer and customer service representatives is acquired. This dialogue text serves as the foundational input data for training and applying the dialogue quality inspection model. Dialogue text can be obtained through various communication channels, such as phone records, online chat logs, or emails.

[0044] Step S202: Obtain the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and the quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship.

[0045] Specifically, the training inference chain corresponding to the dialogue text is obtained. The training inference chain includes at least two levels of quality check prompts. The training inference chain may also include quality check labels for each level of quality check prompts.

[0046] This application employs Enhanced Generative Question Answering (EGQA), a quality check cues-based method. The quality check cues are prompts, which are input prompts used to guide generative models (such as Large Language Models (LLMs)) to generate specific outputs. They provide the model with a clear task or question, enabling it to produce task-related answers or results.

[0047] This application establishes multiple levels of quality control prompts, with at least one prompt at each level. Each quality control prompt is a pre-defined checkpoint designed to evaluate specific elements or behaviors within the dialogue. For example, in an insurance company's customer service system, a customer can book a doctor's appointment through customer service after an accident; for this dialogue text, a quality control checkpoint could be to detect whether the customer service representative has booked a hospital, and the prompt could be "Has a hospital been booked, and what is the name of the hospital?".

[0048] There is a progressive reasoning relationship between the quality inspection prompts. The next level of prompt builds upon the previous level for a more in-depth examination, thus forming a reasoning chain. This chain guides the model to progressively understand and analyze the dialogue content. Each prompt represents a reasoning step, helping the model extract basic information (e.g., appointment information), understand context and details (e.g., confirming appointment date and time), and perform comprehensive analysis and evaluation (finally integrating all extracted information to conduct a quality inspection evaluation of the entire dialogue). This enables the model to gain a deeper understanding of complex dialogues and contextual information, achieving more accurate quality inspection.

[0049] Each quality inspection prompt has a corresponding quality inspection label. The quality inspection label provides the expected answer or standard for the quality inspection point corresponding to the quality inspection prompt.

[0050] Step S203: Based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain the dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results.

[0051] Specifically, the initial dialogue quality inspection model is trained using dialogue text and a training inference chain. This initial dialogue quality inspection model can be a large language model (LLM). A large language model is a deep learning-based natural language processing model that, through pre-training, learns the syntax and semantics of natural language, possesses powerful natural language understanding and reasoning capabilities, and generates human-readable text. For example, the initial dialogue quality inspection model can be built based on the chatg l m3-7B model, which is a model in the ChatGLM3 series and has a parameter scale of up to 7 billion.

[0052] During training, quality control cues of varying levels are gradually introduced to guide the model in understanding and evaluating various aspects of the dialogue text. This progressive training allows the model to gradually transition from simple quality control tasks to more complex ones, ultimately forming a mature dialogue quality control model. Quality control labels, as labeled data, are used to calculate the model loss during training, and then the model parameters are optimized based on this loss.

[0053] Step S204: Obtain the dialogue text to be inspected between the target customer and the target customer service representative, and input the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service representative.

[0054] Specifically, in the application phase, the dialogue text between the target customer and the target customer service representative is acquired and input into a pre-trained dialogue quality inspection model. The dialogue quality inspection model analyzes the dialogue text and generates dialogue quality inspection results, thereby evaluating the performance of the target customer service representative in the dialogue.

[0055] For example, in one embodiment, the input dialogue text to be inspected is:

[0056] Customer: "I broke my finger, please help me make an appointment with a doctor at Sunshine Hospital."

[0057] Customer service: "Okay, I'm making an appointment for you with an orthopedic doctor at Sunshine Hospital. The specialist appointments are this afternoon and tomorrow afternoon. Which date would you like to book?"

[0058] Customer: "The sooner the better."

[0059] Customer service: "Okay, I'll book you an appointment with an orthopedic specialist for 3 p.m. today."

[0060] The output dialogue quality inspection result can be "pass".

[0061] In this embodiment, the dialogue text between the customer and customer service representative is acquired, along with a training inference chain for the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and corresponding quality inspection labels. Each quality inspection prompt corresponds to a quality inspection point in the dialogue text, and the prompts at each level form a progressive inference relationship. Based on the dialogue text, the quality inspection prompts at each level in the training inference chain, and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain the dialogue quality inspection model. The inference chain enables the model to analyze the dialogue content in depth, and the progressive training method allows the model to not only evaluate the surface-level dialogue quality but also to deeply understand the logic, emotional expression, and service quality of the dialogue. The quality inspection prompts guide the initial dialogue quality inspection model to generate prediction results, prompting the model to generate more accurate and comprehensive quality inspection results, covering everything from basic information checks to complex situational understanding. Acquiring the dialogue text between the target customer and the target customer service representative and inputting it into the dialogue quality inspection model can automatically perform customer service dialogue quality inspection and improve the accuracy of customer service dialogue quality inspection.

[0062] Furthermore, step S203 may include: combining the first-level quality inspection prompts in the training inference chain with the dialogue text to obtain the first-level combined text; inputting the first-level combined text into the initial dialogue quality inspection model to instruct the initial dialogue quality inspection model to generate the first-level prediction result based on the first-level quality inspection prompts; calculating the first-level loss based on the first-level prediction result and its corresponding first-level quality inspection label, and adjusting the initial dialogue quality inspection model based on the first-level loss until the first-level loss meets the training stopping condition, completing the first-level training of the initial dialogue quality inspection model, and obtaining the first-level quality inspection result of the initial dialogue quality inspection model that has completed the first-level training for the first-level combined text; for quality inspection prompts after the first-level quality inspection prompts, combining the quality inspection prompts, dialogue text, and the quality inspection result of the previous level to obtain the combined text; performing progressive iterative training on the initial dialogue quality inspection model that has completed the previous level training based on the combined text, until the last level training is completed, to obtain the dialogue quality inspection model.

[0063] Specifically, the first-level quality inspection prompts in the training inference chain are combined with the dialogue text to form the first-level combined text. This combined text is used to clarify the first step of the quality inspection task, allowing the model to focus on specific quality inspection points, such as the quality of the answer to a specific question.

[0064] The first-level combined text is input into the initial dialogue quality inspection model, which generates first-level prediction results based on the guidance of the quality inspection prompts. The first-level prediction results may be information about whether a specific question was answered correctly, or whether a certain behavior exists.

[0065] Based on the first-level prediction results and the preset first-level quality inspection labels, the first-level loss is calculated. The model loss reflects the gap between the model output and the expected result. The model parameters are adjusted based on the first-level loss, and then training is repeated until the first-level loss meets the training stopping condition; the training stopping condition can be that the first-level loss is less than a preset loss threshold. At this point, the first-level training of the initial dialogue quality inspection model is complete. Then, the first-level combined text is input into the model to obtain the model's quality inspection results for the first-level combined text, which will serve as the basis for subsequent training.

[0066] For each subsequent level of quality inspection prompt after the first level, the prompt is combined with the dialogue text and the quality inspection result from the previous level to obtain a combined text. This combination method progressively expands the model's input information, allowing it to refer to the quality inspection conclusions of the previous stage during further training.

[0067] The newly generated combined text is fed into the initial dialogue quality inspection model, which has already completed the previous level of training. The model then makes predictions based on the new quality inspection prompts. This process continues, iteratively training the model until all levels of quality inspection prompts have been used to train the model, forming the final dialogue quality inspection model.

[0068] This progressive training method allows the model to be adjusted and optimized in a targeted manner at each stage, gradually improving the model's ability to understand complex dialogue scenarios and its quality control accuracy.

[0069] In this embodiment, the analytical capabilities of the dialogue quality inspection model are gradually improved through a combination of multi-level quality inspection prompts and corresponding quality inspection labels. The first-level training ensures that the model can accurately identify and process basic quality inspection points. Through the first-level loss feedback and adjustment of model parameters, the model masters the initial quality inspection task. Subsequent progressive training is carried out on the basis of previous training. Combining the quality inspection results of the previous level and new quality inspection prompts, the model is optimized at a deeper level. Progressive iterative training helps the model understand complex dialogue structures and multi-level quality inspection requirements, improves the comprehensive analytical capabilities of the dialogue quality inspection model, and enhances the accuracy of the quality inspection results.

[0070] Furthermore, the steps for calculating the first-level loss based on the first-level prediction results and their corresponding first-level quality inspection labels may include: when performing classification training based on the first-level quality inspection prompts, calculating the first-level loss using the cross-entropy loss function on the first-level prediction results and their corresponding first-level quality inspection labels; when performing generative training based on the first-level quality inspection prompts, calculating the first-level loss using the negative log-likelihood loss function on the first-level prediction results and their corresponding first-level quality inspection labels; and when performing composite training based on the first-level quality inspection prompts, calculating the first-level loss using the composite loss function on the first-level prediction results and their corresponding first-level quality inspection labels.

[0071] Specifically, the first-level training of the initial dialogue quality inspection model based on the first-level quality inspection prompts can be classification training, generation training, or a combination of both.

[0072] For classification training, quality control prompts guide the initial dialogue quality control model to make classification predictions. This involves classification tasks, such as guiding the model to predict whether the dialogue text contains certain content, whether a task is qualified, whether a question has been answered correctly, or whether a specific behavior exists. In this case, the cross-entropy loss function is used to calculate the difference between the predicted result and the quality control label. The cross-entropy loss function is suitable for multi-class classification problems; by measuring the difference between the predicted distribution and the true distribution, it helps the model optimize its classification ability.

[0073] For generation training, where quality control prompts require the model to generate text content, such as summarizing key points of a dialogue or restating a customer's question, the negative log-likelihood (NLL) loss function is used to calculate the model loss. This loss function is suitable for generation tasks, improving the model's generation accuracy by maximizing the probability of the generated text sequences.

[0074] For composite training, the quality control prompts may simultaneously include classification and generation tasks, such as judging the correctness of the answer and the fluency of the language. In this case, a composite loss function is used, which combines the cross-entropy loss and the negative log-likelihood loss. The model loss is obtained by weighted summation of the cross-entropy loss and the negative log-likelihood loss, so as to optimize the objectives of classification and generation at the same time.

[0075] It is understandable that regardless of the training level, it can be classification training, generative training, or composite training, and the model loss calculation method of this embodiment can be used in all of these cases.

[0076] In this embodiment, different types of loss functions are introduced to optimize model training, effectively improving the model's performance on various tasks. When performing classification training, the cross-entropy loss function can improve the model's classification accuracy for different quality checkpoints. When performing generative training, the negative log-likelihood loss function helps the model generate text content that better meets expectations. When performing composite training, the introduction of composite loss functions ensures that the model can optimize the performance of various aspects in a balanced way. The diverse loss calculation methods ensure the accuracy of loss calculation, enhance the model's adaptability, and improve its overall performance in diverse tasks.

[0077] Furthermore, the steps described above for obtaining the dialogue text to be examined between the target customer and the target customer service representative may include: obtaining the voice recording of the dialogue between the target customer and the target customer service representative; performing speech recognition on the voice recording to obtain the initial dialogue text; performing sentiment analysis on the voice recording using a sentiment analysis model to obtain sentiment tags; and combining the initial dialogue text and the sentiment tags to obtain the dialogue text to be examined, with the sentiment tags used to assist in detecting customer satisfaction.

[0078] Specifically, this application can be used in a voice quality inspection system. During the application phase, it obtains the voice recordings from actual conversations between target customers and target customer service representatives, including the full content and intonation of the dialogue.

[0079] Automatic speech recognition (ASR) technology is used to convert spoken dialogue into text format, generating initial dialogue text.

[0080] Building upon dialogue speech recognition, this application uses a sentiment analysis model to analyze the emotional features of the speech. The sentiment analysis model, constructed based on neural networks, can identify the emotions of target customers and target customer service representatives, such as pleasure, anger, and anxiety, based on the audio features of the dialogue speech, and generate emotion labels. In this application, both target customers and target customer service representatives can be assigned emotion labels.

[0081] The initial dialogue text is combined with corresponding emotion tags to generate the dialogue text to be inspected. This not only preserves the dialogue content but also introduces emotional information, providing more multi-dimensional data support for subsequent quality inspection. Emotion tags play an auxiliary role in detecting customer satisfaction, helping to more comprehensively understand the emotions and attitudes of target customers.

[0082] In the application, a corresponding emotion tag can be added to each sentence or several sentences in the dialogue, so as to reflect the emotional changes of each speaker in the dialogue in detail; or one or several overall emotion tags can be added to the entire dialogue, so as to summarize the overall emotional state of the dialogue.

[0083] In this embodiment, the voice dialogue between the target customer and the target customer service representative is acquired, and the initial dialogue text is efficiently generated through speech recognition. The voice dialogue is then analyzed using a sentiment analysis model to obtain sentiment tags, which provide additional emotional information. The sentiment tags can reveal the emotional changes of the customer and customer service representative during the dialogue, and can help detect customer satisfaction. The initial dialogue text and the sentiment tags are combined to obtain the dialogue text to be inspected, so that the accuracy of the dialogue content and the changes in customer emotions can be evaluated simultaneously. This multi-dimensional quality inspection method enhances the accuracy of dialogue quality inspection.

[0084] Furthermore, the steps described above for inputting the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue quality inspection results of the target customer service representative may include: identifying the dialogue type of the dialogue text to be inspected; obtaining the inference chain corresponding to the dialogue type, wherein the inference chain contains at least two levels of quality inspection prompts, the quality inspection prompts correspond to the quality inspection points of the dialogue text to be inspected, and the quality inspection prompts at each level are in a progressive inference relationship; inputting the dialogue text to be inspected and the inference chain into the dialogue quality inspection model for progressive inference to obtain the dialogue quality inspection results of the target customer service representative.

[0085] Specifically, the dialogue text to be examined is analyzed to identify the dialogue type. Dialogue type can be a classification result obtained by categorizing dialogues according to characteristics such as content, purpose, and theme. For example, categorized by content, dialogue types can include technical support (troubleshooting, usage assistance, etc. related to products or services), complaints and suggestions (customers expressing dissatisfaction or suggestions), information inquiries (customers asking about product information, terms and conditions, etc.), and after-sales service (related to returns, exchanges, warranties, etc.).

[0086] Obtain the reasoning chain corresponding to the dialogue type. The reasoning chain contains at least two levels of quality check prompts, with at least one prompt at each level. Each prompt corresponds to a quality check point, i.e., a specific aspect that needs to be checked during quality inspection. The quality check prompts at each level are arranged according to a progressive reasoning relationship. The quality check prompts at the next level will conduct a more in-depth check based on the quality check prompts at the previous level, which means that the understanding and analysis of the dialogue content will gradually deepen.

[0087] The dialogue text to be inspected and the corresponding inference chain are input into the dialogue quality inspection model. The model performs progressive inference according to the quality inspection prompts at each level of the inference chain, gradually checking different aspects of the dialogue, such as the accuracy of information, the resolution of customer problems, and service attitude, until finally generating the dialogue quality inspection result. The inference result at each level provides a deeper contextual understanding for the next step, ensuring the comprehensiveness and accuracy of the quality inspection result.

[0088] In this embodiment, the dialogue type of the dialogue text to be inspected is identified, and the inference chain corresponding to the dialogue type is selected, avoiding generalization and inaccuracy in the quality inspection process. The inference chain contains at least two levels of quality inspection prompts, which correspond to the quality inspection points of the dialogue text to be inspected. Moreover, the quality inspection prompts at each level are in a progressive inference relationship, which enables the dialogue quality inspection model to perform progressive inference based on the dialogue text to be inspected and the inference chain, and to gradually delve into the dialogue content. Each step of the inference is based on the dialogue text to be inspected and the inference result of the previous level, ensuring the accuracy of the quality inspection results.

[0089] Furthermore, the steps described above for identifying the dialogue type of the dialogue text to be inspected may include: inputting the dialogue text to be inspected into a dialogue classification model to obtain the dialogue type of the dialogue text to be inspected, wherein the dialogue classification model performs dialogue classification based on keyword matching or topic modeling; or, inputting the dialogue text to be inspected into a dialogue quality inspection model to obtain the dialogue type of the dialogue text to be inspected.

[0090] Specifically, the dialogue text to be examined is input into a dialogue classification model. This model can classify dialogues using keyword matching or topic modeling methods and output the dialogue type of the dialogue text to be examined.

[0091] Keyword matching involves creating a keyword list for each dialogue type. The keywords in the list reflect the core content of each dialogue type; for example, keywords for a "complaint" type might include "complaint," "problem," and "negative review." The dialogue text to be examined is matched against each keyword list, and the match score is calculated. The match score can be weighted based on factors such as keyword frequency and position. The dialogue text is then categorized into the dialogue type with the highest match score. If multiple categories show similar match scores, a threshold can be set or multiple types can be labeled.

[0092] For topic modeling, preprocessing operations such as word segmentation, stop word removal, and stemming are required on the dialogue text to be examined. Then, topic modeling techniques, such as Latent Dirichlet Allocation (LDA), are used to train the model on the preprocessed dialogue text. The model will automatically identify the topics in the dialogue text and output the corresponding dialogue type.

[0093] As mentioned earlier, dialogue quality inspection models can be large-scale language models (LLMs) with powerful natural language processing capabilities and versatility. They can perform both dialogue quality inspection and dialogue classification tasks. Furthermore, the dialogue text to be inspected can be input into the dialogue quality inspection model to obtain the dialogue type.

[0094] In this embodiment, the dialogue text to be inspected can be classified in various ways using a dialogue classification model or a dialogue quality inspection model. The dialogue classification model classifies dialogues based on keyword matching or topic modeling, ensuring the speed and accuracy of dialogue classification. Inputting the dialogue text to be inspected into the dialogue quality inspection model for dialogue classification simplifies the system architecture, reduces the number of models, and improves processing efficiency.

[0095] Furthermore, the steps described above, which involve inputting the dialogue text to be inspected and the inference chain into the dialogue quality inspection model for progressive inference to obtain the dialogue quality inspection result of the target customer service representative, may include: combining the first-level quality inspection prompt in the inference chain with the dialogue text to be inspected to obtain the first-level text to be inspected; inputting the first-level text to be inspected into the dialogue quality inspection model to instruct the model to generate the first-level quality inspection result based on the first-level quality inspection prompt; and for quality inspection prompts following the first-level prompt, performing progressive inference through the dialogue quality inspection model based on the quality inspection prompt, the dialogue text to be inspected, and the quality inspection result of the previous level, until the inference chain is completed, thereby obtaining the dialogue quality inspection result of the target customer service representative. The dialogue quality inspection result includes the customer service dialogue quality of the target customer service representative and the customer satisfaction of the target customer.

[0096] Specifically, the first-level quality inspection prompts in the inference chain are combined with the dialogue text to be inspected to obtain the first-level text to be inspected. The first-level text to be inspected is then input into the dialogue quality inspection model, and the model generates the first-level quality inspection results based on the first-level quality inspection prompts.

[0097] For each subsequent quality control prompt in the inference chain, progressive reasoning is performed level by level. Each level of reasoning not only considers the new quality control prompt, but also combines the quality control results of the previous level with the dialogue text to be inspected, enabling the model to gradually accumulate an understanding of the dialogue and thus evaluate more complex quality factors.

[0098] A series of progressive reasoning processes will complete the quality inspection and evaluation of the entire reasoning chain, generating comprehensive dialogue quality inspection results for the target customer service representative. These results include the overall quality of the dialogue with the target customer service representative and the target customer's satisfaction, providing a multi-dimensional evaluation.

[0099] For example, consider the following dialogue text to be examined:

[0100] Customer: "I broke my finger, please help me make an appointment with a doctor at Sunshine Hospital."

[0101] Customer service: "Okay, I'm making an appointment for you with an orthopedic doctor at Sunshine Hospital. The specialist appointments are this afternoon and tomorrow afternoon. Which date would you like to book?"

[0102] Customer: "The sooner the better."

[0103] Customer service: "Okay, I'll book you an appointment with an orthopedic specialist for 3 p.m. today."

[0104] The first-level quality inspection prompt is: "Based on the text, answer the following questions: 1. Has a hospital appointment been made? What is the hospital's name? 2. Does the appointment date include information? What is the appointment date?" Combining the first-level quality inspection prompt with the dialogue text to be inspected and inputting it into the dialogue quality inspection model yields the following first-level quality inspection results: "1. Contains an appointment with a hospital; the appointment is with Sunshine Hospital. 2. Contains an appointment time; the appointment time is 3 PM today."

[0105] The dialogue text to be inspected, the first-level quality inspection result, and the second-level quality inspection prompt "Determine whether the dialogue content is qualified from the following questions and answers and analysis, and state the reason" are combined, and then input into the dialogue quality inspection model to obtain the second-level quality inspection result: "Qualified, because the hospital and appointment time are included in the dialogue." Then, the dialogue quality inspection result for the target customer service is generated based on the first-level and second-level quality inspection results.

[0106] In this embodiment, the combination of quality inspection prompts at each level with the dialogue text to be inspected enables the model to analyze the dialogue content in a progressively deeper manner, from simple to complex. The progressive reasoning not only ensures the basic inspection of customer service dialogue quality, but also allows for in-depth exploration of more complex factors such as the logic of the dialogue and the emotional reactions of customers. The cumulative approach enhances the overall analytical capability of the dialogue quality inspection model and improves the accuracy and comprehensiveness of the dialogue quality inspection results.

[0107] It is understandable that the dialogue text can contain emotion tags during the model training phase. Furthermore, the training phase also requires training an emotion recognition model, a dialogue classification model, or an initial dialogue quality control model for dialogue classification; these details will not be elaborated upon here.

[0108] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned dialogue text and the dialogue text to be inspected, the aforementioned dialogue text and the dialogue text to be inspected can also be stored in a node of a blockchain.

[0109] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0110] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0111] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0112] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0113] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0114] Further reference Figure 3 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of a customer service dialogue quality inspection device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0115] like Figure 3As shown, the customer service dialogue quality inspection device 300 described in this embodiment includes: a text acquisition module 301, a training acquisition module 302, a model training module 303, and a dialogue quality inspection module 304, wherein:

[0116] The text acquisition module 301 is used to acquire the dialogue text between the customer and the customer service representative.

[0117] The training acquisition module 302 is used to acquire the training inference chain of the dialogue text. The training inference chain contains at least two levels of quality inspection prompts and the quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship.

[0118] The model training module 303 is used to progressively train the initial dialogue quality inspection model based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels to obtain the dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results.

[0119] The dialogue quality inspection module 304 is used to obtain the dialogue text to be inspected between the target customer and the target customer service representative, and input the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue quality inspection results of the target customer service representative.

[0120] In this embodiment, the dialogue text between the customer and customer service representative is acquired, along with a training inference chain for the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and corresponding quality inspection labels. Each quality inspection prompt corresponds to a quality inspection point in the dialogue text, and the prompts at each level form a progressive inference relationship. Based on the dialogue text, the quality inspection prompts at each level in the training inference chain, and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain the dialogue quality inspection model. The inference chain enables the model to analyze the dialogue content in depth, and the progressive training method allows the model to not only evaluate the surface-level dialogue quality but also to deeply understand the logic, emotional expression, and service quality of the dialogue. The quality inspection prompts guide the initial dialogue quality inspection model to generate prediction results, prompting the model to generate more accurate and comprehensive quality inspection results, covering everything from basic information checks to complex situational understanding. Acquiring the dialogue text between the target customer and the target customer service representative and inputting it into the dialogue quality inspection model can automatically perform customer service dialogue quality inspection and improve the accuracy of customer service dialogue quality inspection.

[0121] In some optional implementations of this embodiment, the model training module 303 may include: a first combination submodule, a first input submodule, a first training submodule, a text combination submodule, and an iterative training submodule, wherein:

[0122] The first combination submodule is used to combine the first-level quality inspection prompts in the training inference chain with the dialogue text to obtain the first-level combined text.

[0123] The first input submodule is used to input the first-level combined text into the initial dialogue quality inspection model, so as to instruct the initial dialogue quality inspection model to generate the first-level prediction result based on the first-level quality inspection prompts.

[0124] The first training submodule is used to calculate the first-level loss based on the first-level prediction results and their corresponding first-level quality inspection labels, and to adjust the initial dialogue quality inspection model based on the first-level loss until the first-level loss meets the training stopping condition, thereby completing the first-level training of the initial dialogue quality inspection model and obtaining the first-level quality inspection results of the first-level combined text by the initial dialogue quality inspection model that has completed the first-level training.

[0125] The text combination submodule is used to combine the quality inspection prompts, dialogue text, and the quality inspection results from the previous level into a combined text for the quality inspection prompts following the first-level quality inspection prompts.

[0126] The iterative training submodule is used to progressively iteratively train the initial dialogue quality inspection model that has completed the previous level of training based on the combined text, until the last level of training is completed, and the dialogue quality inspection model is obtained.

[0127] In this embodiment, the analytical capabilities of the dialogue quality inspection model are gradually improved through a combination of multi-level quality inspection prompts and corresponding quality inspection labels. The first-level training ensures that the model can accurately identify and process basic quality inspection points. Through the first-level loss feedback and adjustment of model parameters, the model masters the initial quality inspection task. Subsequent progressive training is carried out on the basis of previous training. Combining the quality inspection results of the previous level and new quality inspection prompts, the model is optimized at a deeper level. Progressive iterative training helps the model understand complex dialogue structures and multi-level quality inspection requirements, improves the comprehensive analytical capabilities of the dialogue quality inspection model, and enhances the accuracy of the quality inspection results.

[0128] In some optional implementations of this embodiment, the first training submodule may include: a classification calculation unit, a generation calculation unit, and a composite calculation unit, wherein:

[0129] The classification calculation unit is used to calculate the first-level loss by using the cross-entropy loss function on the first-level prediction results and their corresponding first-level quality inspection labels when performing classification training based on the first-level quality inspection prompts.

[0130] The generation calculation unit is used to calculate the first-level loss by using the negative log-likelihood loss function on the first-level prediction results and their corresponding first-level quality inspection labels when generating training based on the first-level quality inspection prompts.

[0131] The composite computing unit is used to calculate the first-level loss by using a composite loss function to calculate the first-level prediction result and its corresponding first-level quality inspection label when performing composite training based on the first-level quality inspection prompts.

[0132] In this embodiment, different types of loss functions are introduced to optimize model training, effectively improving the model's performance on various tasks. When performing classification training, the cross-entropy loss function can improve the model's classification accuracy for different quality checkpoints. When performing generative training, the negative log-likelihood loss function helps the model generate text content that better meets expectations. When performing composite training, the introduction of composite loss functions ensures that the model can optimize the performance of various aspects in a balanced way. The diverse loss calculation methods ensure the accuracy of loss calculation, enhance the model's adaptability, and improve its overall performance in diverse tasks.

[0133] In some optional implementations of this embodiment, the dialogue quality inspection module 304 may include: a voice acquisition submodule, a voice recognition submodule, a sentiment analysis submodule, and a text generation submodule, wherein:

[0134] The voice acquisition submodule is used to acquire the voice conversation between the target customer and the target customer service representative.

[0135] The speech recognition submodule is used to perform speech recognition on the dialogue to obtain the initial dialogue text.

[0136] The sentiment analysis submodule is used to perform sentiment analysis on the dialogue speech using a sentiment analysis model to obtain sentiment labels.

[0137] The text generation submodule is used to combine the initial dialogue text and sentiment tags to obtain the dialogue text to be tested. The sentiment tags are used to assist in the detection of customer satisfaction.

[0138] In this embodiment, the voice dialogue between the target customer and the target customer service representative is acquired, and the initial dialogue text is efficiently generated through speech recognition. The voice dialogue is then analyzed using a sentiment analysis model to obtain sentiment tags, which provide additional emotional information. The sentiment tags can reveal the emotional changes of the customer and customer service representative during the dialogue, and can help detect customer satisfaction. The initial dialogue text and the sentiment tags are combined to obtain the dialogue text to be inspected, so that the accuracy of the dialogue content and the changes in customer emotions can be evaluated simultaneously. This multi-dimensional quality inspection method enhances the accuracy of dialogue quality inspection.

[0139] In some optional implementations of this embodiment, the dialogue quality inspection module 304 may include: a type identification submodule, a reasoning chain acquisition submodule, and a progressive reasoning submodule, wherein:

[0140] The type recognition submodule is used to identify the dialogue type of the dialogue text to be examined.

[0141] The reasoning chain acquisition submodule is used to acquire the reasoning chain corresponding to the dialogue type. The reasoning chain contains at least two levels of quality inspection prompts. The quality inspection prompts correspond to the quality inspection points of the dialogue text to be inspected, and the quality inspection prompts at each level are in a progressive reasoning relationship.

[0142] The progressive reasoning submodule is used to input the dialogue text to be inspected and the reasoning chain into the dialogue quality inspection model for progressive reasoning to obtain the dialogue quality inspection results of the target customer service.

[0143] In this embodiment, the dialogue type of the dialogue text to be inspected is identified, and the inference chain corresponding to the dialogue type is selected, avoiding generalization and inaccuracy in the quality inspection process. The inference chain contains at least two levels of quality inspection prompts, which correspond to the quality inspection points of the dialogue text to be inspected. Moreover, the quality inspection prompts at each level are in a progressive inference relationship, which enables the dialogue quality inspection model to perform progressive inference based on the dialogue text to be inspected and the inference chain, and to gradually delve into the dialogue content. Each step of the inference is based on the dialogue text to be inspected and the inference result of the previous level, ensuring the accuracy of the quality inspection results.

[0144] In some optional implementations of this embodiment, the type identification submodule may include: a first classification unit and a second classification unit, wherein:

[0145] The first classification unit is used to input the dialogue text to be examined into the dialogue classification model to obtain the dialogue type of the dialogue text. The dialogue classification model classifies dialogues based on keyword matching or topic modeling.

[0146] The second classification unit is used to input the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue type of the dialogue text to be inspected.

[0147] In this embodiment, the dialogue text to be inspected can be classified in various ways using a dialogue classification model or a dialogue quality inspection model. The dialogue classification model classifies dialogues based on keyword matching or topic modeling, ensuring the speed and accuracy of dialogue classification. Inputting the dialogue text to be inspected into the dialogue quality inspection model for dialogue classification simplifies the system architecture, reduces the number of models, and improves processing efficiency.

[0148] In some optional implementations of this embodiment, the progressive reasoning submodule may include: a first combination unit, a first input unit, and a progressive reasoning unit, wherein:

[0149] The first combination unit is used to combine the first-level quality inspection prompts in the reasoning chain with the dialogue text to be inspected to obtain the first-level text to be inspected.

[0150] The first input unit is used to input the first-level text to be inspected into the dialogue quality inspection model, so as to instruct the dialogue quality inspection model to generate the first-level quality inspection result according to the first-level quality inspection prompt.

[0151] The progressive reasoning unit is used to perform progressive reasoning on the quality inspection prompts following the first-level quality inspection prompts, based on the quality inspection prompts, the dialogue text to be inspected, and the quality inspection results of the previous level, through the dialogue quality inspection model, until the reasoning is completed according to the reasoning chain, and obtain the dialogue quality inspection results of the target customer service. The dialogue quality inspection results include the customer service dialogue quality of the target customer service and the customer satisfaction of the target customer.

[0152] In this embodiment, the combination of quality inspection prompts at each level with the dialogue text to be inspected enables the model to analyze the dialogue content in a progressively deeper manner, from simple to complex. The progressive reasoning not only ensures the basic inspection of customer service dialogue quality, but also allows for in-depth exploration of more complex factors such as the logic of the dialogue and the emotional reactions of customers. The cumulative approach enhances the overall analytical capability of the dialogue quality inspection model and improves the accuracy and comprehensiveness of the dialogue quality inspection results.

[0153] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed] for details. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0154] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital digital processors (DSPs), embedded devices, etc.

[0155] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0156] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for customer service dialogue quality inspection methods. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0157] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions for the customer service dialogue quality inspection method.

[0158] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0159] The computer device provided in this embodiment can execute the above-described customer service dialogue quality inspection method. The customer service dialogue quality inspection method here can be any of the customer service dialogue quality inspection methods described in the various embodiments above.

[0160] In this embodiment, the dialogue text between the customer and customer service representative is acquired, along with a training inference chain for the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and corresponding quality inspection labels. Each quality inspection prompt corresponds to a quality inspection point in the dialogue text, and the prompts at each level form a progressive inference relationship. Based on the dialogue text, the quality inspection prompts at each level in the training inference chain, and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain the dialogue quality inspection model. The inference chain enables the model to analyze the dialogue content in depth, and the progressive training method allows the model to not only evaluate the surface-level dialogue quality but also to deeply understand the logic, emotional expression, and service quality of the dialogue. The quality inspection prompts guide the initial dialogue quality inspection model to generate prediction results, prompting the model to generate more accurate and comprehensive quality inspection results, covering everything from basic information checks to complex situational understanding. Acquiring the dialogue text between the target customer and the target customer service representative and inputting it into the dialogue quality inspection model can automatically perform customer service dialogue quality inspection and improve the accuracy of customer service dialogue quality inspection.

[0161] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the customer service dialogue quality inspection method described above.

[0162] In this embodiment, the dialogue text between the customer and customer service representative is acquired, along with a training inference chain for the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and corresponding quality inspection labels. Each quality inspection prompt corresponds to a quality inspection point in the dialogue text, and the prompts at each level form a progressive inference relationship. Based on the dialogue text, the quality inspection prompts at each level in the training inference chain, and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain the dialogue quality inspection model. The inference chain enables the model to analyze the dialogue content in depth, and the progressive training method allows the model to not only evaluate the surface-level dialogue quality but also to deeply understand the logic, emotional expression, and service quality of the dialogue. The quality inspection prompts guide the initial dialogue quality inspection model to generate prediction results, prompting the model to generate more accurate and comprehensive quality inspection results, covering everything from basic information checks to complex situational understanding. Acquiring the dialogue text between the target customer and the target customer service representative and inputting it into the dialogue quality inspection model can automatically perform customer service dialogue quality inspection and improve the accuracy of customer service dialogue quality inspection.

[0163] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0164] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for quality inspection of a customer service dialogue, characterized by, Includes the following steps: Obtain the text of the conversation between the customer and customer service; Obtain the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship. Based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels, the initial dialogue quality inspection model is progressively trained to obtain a dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results. Obtain the dialogue text between the target customer and the target customer service representative, and input the dialogue text into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service representative. The step of progressively training the initial dialogue quality inspection model to obtain the dialogue quality inspection model based on the dialogue text, the quality inspection prompts at each level in the training inference chain, and their corresponding quality inspection labels includes: The first-level quality inspection prompt in the training inference chain is combined with the dialogue text to obtain the first-level combined text; The first-level combined text is input into the initial dialogue quality inspection model to instruct the initial dialogue quality inspection model to generate a first-level prediction result based on the first-level quality inspection prompt. The first-level loss is calculated based on the first-level prediction result and its corresponding first-level quality inspection label. The initial dialogue quality inspection model is then adjusted based on the first-level loss until the first-level loss meets the training stopping condition. The first-level training of the initial dialogue quality inspection model is completed, and the first-level quality inspection result of the initial dialogue quality inspection model after completing the first-level training is obtained for the first-level combined text. For quality inspection prompts following the first-level quality inspection prompts, the quality inspection prompts, the dialogue text, and the quality inspection results from the previous level are combined to obtain a combined text; Based on the combined text, the initial dialogue quality inspection model that has completed the previous level of training is progressively iteratively trained until the last level of training is completed, thus obtaining the dialogue quality inspection model.

2. The customer care conversation quality inspection method of claim 1, wherein, The step of calculating the first-level loss based on the first-level prediction result and its corresponding first-level quality inspection label includes: When performing classification training based on the first-level quality inspection prompts, the first-level loss is calculated by using the cross-entropy loss function on the first-level prediction results and their corresponding first-level quality inspection labels. When generating training based on the first-level quality inspection prompts, the first-level loss is calculated by using the negative log-likelihood loss function on the first-level prediction results and their corresponding first-level quality inspection labels. When performing composite training based on the first-level quality inspection prompts, the first-level loss is calculated by using a composite loss function on the first-level prediction results and their corresponding first-level quality inspection labels.

3. The method of claim 1, wherein, The steps for obtaining the text of the dialogue to be examined between the target customer and the target customer service representative include: Obtain audio recordings of conversations between target customers and target customer service representatives; The dialogue speech is subjected to speech recognition to obtain the initial dialogue text; The emotional analysis model is used to perform emotional analysis on the dialogue voice to obtain emotional labels. The initial dialogue text and the emotion tag are combined to obtain the dialogue text to be tested, and the emotion tag is used to assist in the detection of customer satisfaction.

4. The customer service dialogue quality inspection method according to claim 1, characterized in that, The step of inputting the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service includes: Identify the dialogue type of the dialogue text to be examined; Obtain the reasoning chain corresponding to the dialogue type. The reasoning chain contains at least two levels of quality inspection prompts. The quality inspection prompts correspond to the quality inspection points of the dialogue text to be inspected, and the quality inspection prompts at each level are in a progressive reasoning relationship. The dialogue text to be inspected and the inference chain are input into the dialogue quality inspection model for progressive inference to obtain the dialogue quality inspection result of the target customer service.

5. The customer service dialogue quality inspection method according to claim 4, characterized in that, The step of identifying the dialogue type of the dialogue text to be examined includes: The dialogue text to be examined is input into a dialogue classification model to obtain the dialogue type of the dialogue text. The dialogue classification model performs dialogue classification based on keyword matching or topic modeling; or... The dialogue text to be inspected is input into the dialogue quality inspection model to obtain the dialogue type of the dialogue text to be inspected.

6. The customer service dialogue quality inspection method according to claim 4, characterized in that, The step of inputting the dialogue text to be inspected and the inference chain into the dialogue quality inspection model for progressive inference to obtain the dialogue quality inspection result of the target customer service includes: The first-level quality inspection prompt in the inference chain is combined with the dialogue text to be inspected to obtain the first-level text to be inspected; The first-level text to be inspected is input into the dialogue quality inspection model to instruct the dialogue quality inspection model to generate a first-level quality inspection result based on the first-level quality inspection prompt. For quality inspection prompts following the first-level quality inspection prompts, based on the quality inspection prompts, the dialogue text to be inspected, and the quality inspection results of the previous level, progressive reasoning is performed through the dialogue quality inspection model until the reasoning is completed according to the reasoning chain, and the dialogue quality inspection results of the target customer service are obtained. The dialogue quality inspection results include the customer service dialogue quality of the target customer service and the customer satisfaction of the target customer.

7. A customer service dialogue quality inspection device, characterized in that, include: The text acquisition module is used to acquire the dialogue text between the customer and customer service. The training acquisition module is used to acquire the training inference chain of the dialogue text. The training inference chain includes at least two levels of quality inspection prompts and quality inspection labels corresponding to each quality inspection prompt. The quality inspection prompts correspond to the quality inspection points of the dialogue text, and the quality inspection prompts at each level are in a progressive inference relationship. The model training module is used to progressively train the initial dialogue quality inspection model based on the dialogue text, the quality inspection prompts at each level in the training inference chain and their corresponding quality inspection labels to obtain a dialogue quality inspection model. The quality inspection prompts are used to guide the initial dialogue quality inspection model to generate prediction results. The dialogue quality inspection module is used to obtain the dialogue text to be inspected between the target customer and the target customer service, and input the dialogue text to be inspected into the dialogue quality inspection model to obtain the dialogue quality inspection result of the target customer service. The model training module includes: a first combination submodule, a first input submodule, a first training submodule, a text combination submodule, and an iterative training submodule, wherein: The first combination submodule is used to combine the first-level quality inspection prompts in the training inference chain with the dialogue text to obtain the first-level combined text; The first input submodule is used to input the first-level combined text into the initial dialogue quality inspection model, so as to instruct the initial dialogue quality inspection model to generate the first-level prediction result according to the first-level quality inspection prompts; The first training submodule is used to calculate the first-level loss based on the first-level prediction results and their corresponding first-level quality inspection labels, and to adjust the initial dialogue quality inspection model based on the first-level loss until the first-level loss meets the training stopping condition, thereby completing the first-level training of the initial dialogue quality inspection model and obtaining the first-level quality inspection results of the first-level combined text by the initial dialogue quality inspection model that has completed the first-level training. The text combination submodule is used to combine the quality inspection prompts, dialogue text, and the quality inspection results of the previous level into a combined text for the quality inspection prompts following the first-level quality inspection prompts. The iterative training submodule is used to progressively iteratively train the initial dialogue quality inspection model that has completed the previous level of training based on the combined text, until the last level of training is completed, and the dialogue quality inspection model is obtained.

8. A computer device comprising a memory and a processor, the memory storing computer-readable instructions, wherein the processor, when executing the computer-readable instructions, implements the steps of the customer service dialogue quality inspection method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the customer service dialogue quality inspection method as described in any one of claims 1 to 6.

Images