Customer service call quality inspection method and device, computer device and storage medium
By embedding encoding, type recognition, emotion recognition, and tone recognition into customer service calls, and combining attention weights and multi-class hierarchical models, the problem of low accuracy in customer service call quality inspection has been solved, achieving more efficient and accurate quality inspection results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA PING AN PROPERTY INSURANCE CO LTD
- Filing Date
- 2024-01-05
- Publication Date
- 2026-06-23
Smart Images
Figure CN117793251B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of artificial intelligence and financial technology, and in particular to a method, apparatus, computer equipment and storage medium for quality inspection of customer service calls. Background Technology
[0002] In production and daily life, customer service personnel, such as agents, frequently need to communicate with customers to complete the company's tasks or provide services. Companies also need to conduct quality checks on customer service calls to assess whether customer service representatives adhere to relevant regulations and to evaluate service quality. For example, in the financial insurance sector, customer service representatives (such as agents or salespersons at insurance companies) need to contact customers by phone to recommend insurance products and explain service procedures. Insurance companies need to conduct quality checks on customer service calls to ensure that customer service representatives comply with relevant laws and company regulations, and to understand the effectiveness and results of their insurance product recommendations.
[0003] Currently, customer service call quality inspection is usually done manually. Due to the large amount of call data, quality inspectors can only perform random checks. The quality inspection process is affected by personal cognition and experience, which leads to low accuracy in customer service call quality inspection. Summary of the Invention
[0004] The purpose of this application is to provide a customer service call quality inspection method, apparatus, computer equipment, and storage medium to solve the problem of low accuracy in customer service call quality inspection.
[0005] To address the aforementioned technical problems, this application provides a customer service call quality inspection method, employing the following technical solution:
[0006] Obtain customer service call information;
[0007] For each customer service call in the call information, the customer service call is embedded and encoded using a pre-trained model to obtain the call vector of the customer service call;
[0008] The call vectors of each customer service call are identified using a type recognition model to determine the quality inspection type involved in the call information.
[0009] The emotional features of the customer service call are obtained through an emotion recognition model, and the tone features of the customer service call are obtained through a tone recognition model.
[0010] Based on the call vector, emotional features, and tone features corresponding to the customer service call, an initial composite vector for the customer service call is generated.
[0011] Attention weights are added to the initial composite vectors of each customer service call to obtain the composite vectors of each customer service call;
[0012] The composite vector of each customer service call is input into the quality inspection point detection model under the quality inspection type to obtain the point detection results output by each quality inspection point detection model, and the call quality inspection result of the customer service is generated based on the point detection results.
[0013] To address the aforementioned technical problems, this application also provides a customer service call quality inspection device, which employs the following technical solution:
[0014] The call acquisition module is used to acquire call information from customer service representatives.
[0015] The call encoding module is used to embed and encode each customer service call in the call information using a pre-trained model to obtain the call vector of the customer service call.
[0016] The type recognition module is used to identify the quality inspection type of each customer service call based on the type recognition model, and obtain the quality inspection type involved in the call information.
[0017] The feature acquisition module is used to acquire the emotional features of the customer service call through an emotion recognition model and to the tone features of the customer service call through a tone recognition model.
[0018] The initial generation module is used to generate an initial composite vector for the customer service call based on the call vector, emotional features, and tone features corresponding to the customer service call.
[0019] The weighting module is used to add attention weights to the initial composite vector of each customer service call to obtain the composite vector of each customer service call.
[0020] The call quality inspection module is used to input the composite vector of each customer service call into the quality inspection point detection model under the quality inspection type, obtain the point detection results output by each quality inspection point detection model, and generate the call quality inspection result of the customer service based on the point detection results.
[0021] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:
[0022] Obtain customer service call information;
[0023] For each customer service call in the call information, the customer service call is embedded and encoded using a pre-trained model to obtain the call vector of the customer service call;
[0024] The call vectors of each customer service call are identified using a type recognition model to determine the quality inspection type involved in the call information.
[0025] The emotional features of the customer service call are obtained through an emotion recognition model, and the tone features of the customer service call are obtained through a tone recognition model.
[0026] Based on the call vector, emotional features, and tone features corresponding to the customer service call, an initial composite vector for the customer service call is generated.
[0027] Attention weights are added to the initial composite vectors of each customer service call to obtain the composite vectors of each customer service call;
[0028] The composite vector of each customer service call is input into the quality inspection point detection model under the quality inspection type to obtain the point detection results output by each quality inspection point detection model, and the call quality inspection result of the customer service is generated based on the point detection results.
[0029] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:
[0030] Obtain customer service call information;
[0031] For each customer service call in the call information, the customer service call is embedded and encoded using a pre-trained model to obtain the call vector of the customer service call;
[0032] The call vectors of each customer service call are identified using a type recognition model to determine the quality inspection type involved in the call information.
[0033] The emotional features of the customer service call are obtained through an emotion recognition model, and the tone features of the customer service call are obtained through a tone recognition model.
[0034] Based on the call vector, emotional features, and tone features corresponding to the customer service call, an initial composite vector for the customer service call is generated.
[0035] Attention weights are added to the initial composite vectors of each customer service call to obtain the composite vectors of each customer service call;
[0036] The composite vector of each customer service call is input into the quality inspection point detection model under the quality inspection type to obtain the point detection results output by each quality inspection point detection model, and the call quality inspection result of the customer service is generated based on the point detection results.
[0037] Compared with existing technologies, the embodiments of this application have the following main advantages: Obtaining customer service call information; for each sentence of customer service call information, embedding and encoding the customer service call using a large-scale natural language pre-trained model can more accurately generate the call vector of the customer service call; performing quality inspection type identification on the call vector of each customer service call according to a type recognition model to determine the quality inspection type involved in the call information; in order to obtain information beyond the literal meaning, obtaining the emotional features of the customer service call through an emotion recognition model to capture the emotional tendency in the customer service call; obtaining the tone features of the customer service call through a tone recognition model to analyze the tone in the customer service call; generating the initial customer service call based on the call vector, emotional features, and tone features corresponding to the customer service call. Composite vectors, integrating multi-dimensional features, improve the accuracy of quality inspection. Attention weights are added to the initial composite vectors of each customer service call to obtain composite vectors for each call, allowing the model to focus more on important calls, further improving accuracy. The composite vectors of each call are input into the detection models for each quality inspection point under each quality inspection type. Each detection model performs binary classification prediction for a specific quality inspection point, accurately outputting the detection results at that point. Based on these results, the call quality inspection results are generated, further improving the accuracy of the call quality inspection results. This application combines multi-dimensional features and uses a multi-class hierarchical model to progressively inspect call information, significantly improving the accuracy and efficiency of customer service call quality inspection. Attached Figure Description
[0038] 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.
[0039] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;
[0040] Figure 2 This is a flowchart of one embodiment of the customer service call quality inspection method according to this application;
[0041] Figure 3 This is a schematic diagram of a structure of an embodiment of the customer service call quality inspection device according to this application;
[0042] Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation
[0043] 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.
[0044] 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.
[0045] 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.
[0046] 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.
[0047] 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.
[0048] 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.
[0049] 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.
[0050] It should be noted that the customer service call quality inspection method provided in this application embodiment is generally executed by the server, and correspondingly, the customer service call quality inspection device is generally set in the server.
[0051] 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.
[0052] Continue to refer to Figure 2 A flowchart of an embodiment of the customer service call quality inspection method according to this application is shown. The customer service call quality inspection method includes the following steps:
[0053] Step S201: Obtain customer service call information.
[0054] In this embodiment, the customer service call 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, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future known wireless connection methods.
[0055] Specifically, this involves obtaining call information from customer service representatives who require call quality inspection. Call information refers to call data generated during voice or text communication between customer service representatives and customers. Typically, call information exists primarily in text format; if the customer service representative communicates with the customer via voice, the voice can be converted to text.
[0056] Furthermore, step S201 above may include: when it is detected that the customer service representative is in a communication state, obtaining the customer service representative's real-time call information through Kafka; or, according to the triggered quality inspection instruction, obtaining the customer service representative's historical call information from the database.
[0057] Specifically, when it is detected that a customer service representative is communicating with a customer, the customer service system can convert the representative's real-time voice into text-based call information and push it to the customer service call quality inspection system through the Kafka middleware for real-time customer service call quality inspection.
[0058] Alternatively, a quality inspection instruction can be triggered periodically. The customer service call quality inspection system can then retrieve historical call information from the database based on the instruction and perform non-real-time customer service call quality inspection.
[0059] In this embodiment, real-time call information from customer service can be obtained through Kafka for real-time call quality inspection, thereby promptly informing customer service of the call quality inspection results or providing reminders; alternatively, based on triggered quality inspection instructions, historical call information from customer service can be retrieved from the database for non-real-time call quality inspection, improving the flexibility of call quality inspection.
[0060] Step S202: For each customer service call in the call information, the customer service call is embedded and encoded using a pre-trained model to obtain the call vector of the customer service call.
[0061] Specifically, since a call is a multi-round process, the call information can contain multiple sentences from the customer service representative.
[0062] Each customer service call is embedded and encoded using a pre-trained model to obtain a call vector, thus achieving a vectorized representation of the text. In one embodiment, the pre-trained model can employ SentenceTransformers, using a large-scale natural language processing pre-trained model to generate call vectors. This significantly reduces the problem of insufficient corpus data, allowing the pre-trained model to better understand and process text data, and accurately represent the features of customer service calls.
[0063] Furthermore, prior to step S201 above, the process may include: obtaining a customer service call quality inspection dataset, which contains quality inspection standard information and quality inspection sample information; and fine-tuning the initial pre-trained model based on the customer service call quality inspection dataset to obtain a pre-trained model.
[0064] Specifically, this application also requires pre-tuning of the pre-trained model: obtaining a customer service call quality inspection dataset, which contains quality inspection standard information and quality inspection sample information. The quality inspection standard information can include laws and regulations, company regulations (for example, prohibiting threatening or abusing customers, requiring the use of polite language, and prohibiting forcing insurance companies to underwrite policies, all of which fall under company regulations), and company requirements (more business-specific, such as requiring customer service representatives to recommend certain insurance products using prescribed scripts). The quality inspection sample information consists of actual call messages with labels; these labels are used to annotate specific domains, such as insurance in the insurance company's quality inspection sample information; the labels can also indicate the type of quality inspection involved in the sample (laws and regulations, company regulations, and company requirements), and whether it meets specific quality inspection points (e.g., whether a certain call message threatened the customer).
[0065] The initial pre-trained model is fine-tuned using a customer service call quality inspection dataset to obtain a pre-trained model. Fine-tuning with a customer service call quality inspection dataset specific to a particular domain (e.g., the auto insurance industry) allows the pre-trained model to better adapt to the characteristics and content of that domain. The pre-trained model gradually learns domain-specific terminology, expressions, and compliance regulations. This fine-tuning process enables the pre-trained model to better capture domain-specific textual features, thereby generating more accurate vectorized representations of call information.
[0066] In this embodiment, a customer service call quality inspection dataset is obtained, which contains quality inspection standard information and quality inspection sample information. The initial pre-trained model is fine-tuned based on the customer service call quality inspection dataset to obtain a pre-trained model. The pre-trained model is better adapted to the characteristics and content of the specific domain and can more accurately generate vectorized representations for call information.
[0067] Step S203: Based on the type recognition model, perform quality inspection type identification on the call vectors of each customer service call to obtain the quality inspection type involved in the call information.
[0068] Specifically, this application establishes a multi-class hierarchical model, which is a composite model composed of multiple models. The first level of the multi-class hierarchical model is a type recognition model, which can be constructed from an artificial intelligence-based model.
[0069] The call vectors of each customer service call are used as a whole to input the type recognition model for quality inspection type recognition. It can identify which quality inspection types the call information involves (i.e., the laws and regulations, corporate regulations and corporate requirements mentioned above, i.e., quality inspection type recognition determines which quality inspection category the call information involves) and output the quality inspection types involved in the call information.
[0070] Step S204: Obtain the emotional features of the customer service call through the emotion recognition model, and obtain the tone features of the customer service call through the tone recognition model.
[0071] Specifically, due to the complexity of language, customer service calls may also contain information beyond the literal meaning. This application categorizes this information beyond the literal meaning as emotion and tone, which can cause customer service calls to have meanings beyond the literal meaning. For example, the idiom "sincerity" is an expression of this situation.
[0072] To this end, this application also inputs customer service calls into an emotion recognition model to capture the emotional tendencies in the calls, such as positive, negative, and neutral, and outputs emotional features representing these tendencies, such as probability values representing the most likely emotion. The application also inputs customer service calls into the emotion recognition model to analyze the tone of voice, such as sarcasm, seriousness, and anger, and outputs tone features representing these tones.
[0073] Emotion recognition models and tone recognition models can also be built from artificial intelligence-based models.
[0074] Step S205: Generate an initial composite vector for the customer service call based on the call vector, emotional features, and tone features corresponding to the customer service call.
[0075] Specifically, in order to accurately conduct customer service call quality inspection, this application generates an initial composite vector for the customer service call based on the call vector, emotional features, and tone features corresponding to the customer service call, so as to integrate the call vector, emotional features, and tone features.
[0076] Furthermore, step S205 may include: concatenating the call vector, emotional features, and tone features corresponding to the customer service call in a preset order to obtain the initial composite vector of the customer service call.
[0077] Specifically, the call vector, emotion feature, and tone feature corresponding to the customer service call are concatenated in a preset order to obtain the initial composite vector of the customer service call. In one embodiment, the initial composite vector can be obtained by concatenating the vectors in the order of "call vector - emotion feature - tone feature".
[0078] In this embodiment, the call vector, emotional features, and tone features corresponding to the customer service call are concatenated in a preset order to accurately and orderly obtain the initial composite vector of each customer service call.
[0079] Step S206: Add attention weights to the initial composite vector of each customer service call to obtain the composite vector of each customer service call.
[0080] Specifically, the semantic value and importance of different customer service calls vary. For example, a customer service call where the customer service representative greets the customer and introduces themselves, or a customer service call where the customer service representative recommends a product, clearly has higher semantic value and importance.
[0081] The customer service call quality inspection system proposes an attention mechanism. This mechanism adds attention weights to the initial composite vectors of each customer service call, resulting in composite vectors for each call. The more important the customer service call, the greater its attention weight, and the more attention the quality inspection model will pay to these calls.
[0082] Step S207: Input the composite vector of each customer service call into the detection model of each quality inspection point under the quality inspection type, obtain the point detection results output by each quality inspection point detection model, and generate the call quality inspection results of the customer service based on the point detection results.
[0083] Specifically, the second level of the multi-class hierarchical model is the point detection model, which is used to detect whether customer service has violated a specific quality control point, such as whether a certain call message threatened or insulted the customer.
[0084] As mentioned above, the type recognition model identifies the quality inspection type, determining which quality inspection categories the call information falls under (specifically, which major category of quality inspection it pertains to within laws, regulations, company rules, and requirements). Each quality inspection type includes multiple quality inspection point detection models, each used to detect whether customer service representatives have violated the corresponding quality inspection point. For example, under the quality inspection type of "company rules," models such as threat detection, verbal abuse detection, and coercion detection could be included. Each detection model can be a binary classification model, separately detecting whether the customer service representative threatened, verbally abused, or coerced the customer into doing certain things.
[0085] The composite vector of each customer service call is treated as a whole and input into the detection model of each quality inspection point under the quality inspection type. The point detection results output by each quality inspection point detection model are obtained, and the call quality inspection results of the customer service are generated by combining the detection results of each point.
[0086] In this embodiment, customer service call information is obtained. For each sentence of the customer service call in the call information, the call is embedded and encoded using a large-scale natural language pre-trained model to more accurately generate a call vector. A type recognition model is used to perform quality inspection type identification on the call vectors of each customer service call to determine the quality inspection type involved in the call information. To obtain information beyond the literal meaning, an emotion recognition model is used to obtain the emotional features of the customer service call to capture the emotional tendency in the call. A tone recognition model is used to obtain the tone features of the customer service call to analyze the tone in the call. Based on the call vector, emotion features, and tone features corresponding to the customer service call, an initial composite vector of the customer service call is generated, integrating multiple... The application incorporates multi-dimensional features to improve the accuracy of quality inspection. Attention weights are added to the initial composite vectors of each customer service call to obtain composite vectors for each call, allowing the model to focus more on important calls, further enhancing accuracy. The composite vectors of each call are input into the detection models for each quality inspection point under each quality inspection type. Each detection model performs binary classification prediction for a specific quality inspection point, accurately outputting the detection results at that point. Based on these results, the call quality inspection results are generated, further improving the accuracy of the call quality inspection results. This application combines multi-dimensional features with a multi-class hierarchical model to progressively inspect call information, significantly improving the accuracy and efficiency of customer service call quality inspection.
[0087] Furthermore, step S204 may include: inputting the text-based customer service call into an emotion recognition model to obtain the emotion features of the customer service call, and inputting the text-based customer service call into a tone recognition model to obtain the tone features of the customer service call; or, extracting the acoustic features of the voice-based customer service call, and extracting initial emotion features and initial tone features from the acoustic features; inputting the initial emotion features into an emotion recognition model to obtain the emotion features of the customer service call, and inputting the initial tone features into a tone recognition model to obtain the tone features of the customer service call.
[0088] Specifically, regardless of whether the customer service call is in text or voice format, the customer service call quality inspection system can extract emotional and tone features.
[0089] If the customer service call is in text format, the call is input into the emotion recognition model to obtain the emotion features of the call, and the call is input into the tone recognition model to obtain the tone features of the call.
[0090] If the call information input into the customer service call quality inspection system is in voice form, then the acoustic features of the customer service call are extracted, such as MFCC (Mel-frequency cepstral coefficients), tone, energy, fundamental frequency, zero-crossing rate, etc.
[0091] Before extracting acoustic features, customer service calls can be preprocessed, including noise removal and audio quality standardization, to ensure signal accuracy and clarity.
[0092] Then, initial sentiment features and initial tone features are extracted from the acoustic features. When extracting initial sentiment features, features related to sentiment analysis can be combined with the acoustic features. For example, customer service calls can be divided into smaller time segments, and features such as sound intensity and pitch changes within each segment can be calculated to extract potentially sentiment-related information. When extracting initial tone features, features related to tone analysis can be combined with the acoustic features. Initial tone features may include pitch fluctuations, changes in speech rate, and changes in volume; these features may be related to the intensity and intensity of the tone.
[0093] Finally, the initial emotional features are input into the emotional recognition model to obtain the emotional features of the customer service call; the initial tone features are input into the tone recognition model to obtain the tone features of the customer service call.
[0094] Both emotion recognition models and tone recognition models require pre-training. They can both employ deep learning models, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), to associate acoustic features with emotion labels.
[0095] It is understandable that both the type recognition model and the detection models for each quality checkpoint need to be trained in advance, such as through supervised training.
[0096] In this embodiment, emotional features / tone features can be directly extracted from customer service calls in text form through an emotion recognition model / tone recognition model; or, acoustic features can be extracted from customer service calls in voice form, and initial emotional features and initial tone features can be further extracted and input into the emotion recognition model and tone recognition model respectively to obtain emotional features and tone features, thus realizing the acquisition of emotional features and tone features in various situations.
[0097] Furthermore, step S206 may include: for each customer service call, calculating the attention score between the customer service call and other customer service calls; calculating the average attention score of the customer service call based on the attention score between the customer service call and other customer service calls; normalizing the average attention score of each customer service call to obtain the attention weight of each customer service call; and adding the attention weight of each customer service call to the initial composite vector of each customer service call to obtain the composite vector of each customer service call.
[0098] Specifically, for each customer service call, an attention score is calculated between that customer service call and all other customer service calls. The attention score reflects the similarity or importance between different customer service calls.
[0099] The average attention score of this customer service call is calculated by averaging its attention scores with those of all other customer service calls. For each customer service call, its average attention score can be calculated. The calculated attention scores are then normalized (e.g., using Softmax), and the normalized values are used as the attention weights for each customer service call. The sum of the attention weights for all customer service calls is 1.
[0100] The average attention score reflects the average similarity or correlation between customer service calls and other customer service calls. It can be understood that if the average correlation between a sentence and other sentences is higher, it means that the sentence is more closely related to other sentences and therefore needs more attention, thus corresponding to a greater attention weight.
[0101] The attention weights of each customer service call are added to the initial composite vector of each customer service call to obtain the composite vector of each customer service call.
[0102] In this embodiment, for each customer service call, the attention score between the customer service call and other customer service calls is calculated. Based on the attention scores between the customer service call and other customer service calls, the average attention score of the customer service call is calculated. The attention score reflects the similarity or importance of the customer service call to other customer service calls, while the average attention score reflects the average similarity or correlation between the customer service call and other customer service calls. The stronger the correlation, the more attention should be paid. After normalizing the average attention score, the resulting attention weight is also higher. The attention weight of each customer service call is added to the initial composite vector of each customer service call to obtain the composite vector of each customer service call. Adding the attention weight to the vector ensures that the subsequent customer service call quality inspection can be more accurate.
[0103] Furthermore, the above steps for calculating the attention score between each customer service call and all other customer service calls are as follows: For each customer service call, calculate the similarity between the call vector of the customer service call and the call vectors of all other customer service calls, and use the obtained similarity scores as the attention score between the customer service call and all other customer service calls.
[0104] Specifically, for each customer service call, the similarity between the call vector of the customer service call and the call vectors of other customer service calls is calculated, for example, by calculating the cosine similarity, and the obtained similarity is used as the attention score between the customer service call and other customer service calls.
[0105] In this embodiment, the similarity between the call vector of a customer service call and the call vectors of other customer service calls is calculated, and the obtained similarity is used as the attention score between the customer service call and other customer service calls, thus realizing the calculation of attention score and the introduction of attention.
[0106] It should be emphasized that, to further ensure the privacy and security of the aforementioned call information, the call information can also be stored in a blockchain node.
[0107] 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.
[0108] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use 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.
[0109] 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.
[0110] 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 with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. 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).
[0111] 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.
[0112] 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 call 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.
[0113] like Figure 3As shown, the customer service call quality inspection device 300 described in this embodiment includes: a call acquisition module 301, a call encoding module 302, a type recognition module 303, a feature acquisition module 304, an initial generation module 305, a weight addition module 306, and a call quality inspection module 307, wherein:
[0114] The call acquisition module 301 is used to acquire call information from customer service representatives.
[0115] The call encoding module 302 is used to embed and encode each customer service call in the call information using a pre-trained model to obtain the call vector of the customer service call.
[0116] The type recognition module 303 is used to perform quality inspection type recognition on the call vector of each customer service call according to the type recognition model, and obtain the quality inspection type involved in the call information.
[0117] The feature acquisition module 304 is used to acquire the emotional features of customer service calls through the emotion recognition model and to the tone features of customer service calls through the tone recognition model.
[0118] The initial generation module 305 is used to generate an initial composite vector for a customer service call based on the call vector, emotional features, and tone features corresponding to the customer service call.
[0119] The weighting module 306 is used to add attention weights to the initial composite vector of each customer service call to obtain the composite vector of each customer service call.
[0120] The call quality inspection module 307 is used to input the composite vector of each customer service call into the detection model of each quality inspection point under the quality inspection type, obtain the point detection results output by each quality inspection point detection model, and generate the call quality inspection results of the customer service based on the detection results of each point.
[0121] In this embodiment, customer service call information is obtained. For each sentence of the customer service call in the call information, the call is embedded and encoded using a large-scale natural language pre-trained model to more accurately generate a call vector. A type recognition model is used to perform quality inspection type identification on the call vectors of each customer service call to determine the quality inspection type involved in the call information. To obtain information beyond the literal meaning, an emotion recognition model is used to obtain the emotional features of the customer service call to capture the emotional tendency in the call. A tone recognition model is used to obtain the tone features of the customer service call to analyze the tone in the call. Based on the call vector, emotion features, and tone features corresponding to the customer service call, an initial composite vector of the customer service call is generated, integrating multiple... The application incorporates multi-dimensional features to improve the accuracy of quality inspection. Attention weights are added to the initial composite vectors of each customer service call to obtain composite vectors for each call, allowing the model to focus more on important calls, further enhancing accuracy. The composite vectors of each call are input into the detection models for each quality inspection point under each quality inspection type. Each detection model performs binary classification prediction for a specific quality inspection point, accurately outputting the detection results at that point. Based on these results, the call quality inspection results are generated, further improving the accuracy of the call quality inspection results. This application combines multi-dimensional features with a multi-class hierarchical model to progressively inspect call information, significantly improving the accuracy and efficiency of customer service call quality inspection.
[0122] In some optional implementations of this embodiment, the customer service call quality inspection device 300 may further include: a dataset acquisition module and a pre-training module, wherein:
[0123] The dataset acquisition module is used to acquire customer service call quality inspection datasets, which contain quality inspection standard information and quality inspection sample information.
[0124] The pre-training module is used to fine-tune the initial pre-trained model based on the customer service call quality inspection dataset to obtain the pre-trained model.
[0125] In this embodiment, a customer service call quality inspection dataset is obtained, which contains quality inspection standard information and quality inspection sample information. The initial pre-trained model is fine-tuned based on the customer service call quality inspection dataset to obtain a pre-trained model. The pre-trained model is better adapted to the characteristics and content of the specific domain and can more accurately generate vectorized representations for call information.
[0126] In some optional implementations of this embodiment, the call acquisition module 301 may include: a real-time acquisition submodule and a historical acquisition submodule, wherein:
[0127] The real-time acquisition submodule is used to obtain real-time call information from customer service representatives via Kafka when a call is detected to be in progress.
[0128] The history retrieval submodule is used to retrieve historical call information of customer service from the database based on the triggered quality inspection instructions.
[0129] In this embodiment, real-time call information from customer service can be obtained through Kafka for real-time call quality inspection, thereby promptly informing customer service of the call quality inspection results or providing reminders; alternatively, based on triggered quality inspection instructions, historical call information from customer service can be retrieved from the database for non-real-time call quality inspection, improving the flexibility of call quality inspection.
[0130] In some optional implementations of this embodiment, the feature acquisition module 304 may include: a text acquisition submodule, a feature extraction submodule, and a feature input submodule, wherein:
[0131] The text acquisition submodule is used to input the text form of customer service calls into the emotion recognition model to obtain the emotion features of the customer service calls, and to input the text form of customer service calls into the tone recognition model to obtain the tone features of the customer service calls.
[0132] The feature extraction submodule is used to extract the acoustic features of customer service calls in voice form, and extract initial emotion features and initial tone features from the acoustic features.
[0133] The feature input submodule is used to input the initial emotional features into the emotional recognition model to obtain the emotional features of the customer service call, and to input the initial tone features into the tone recognition model to obtain the tone features of the customer service call.
[0134] In this embodiment, emotional features / tone features can be directly extracted from customer service calls in text form through an emotion recognition model / tone recognition model; or, acoustic features can be extracted from customer service calls in voice form, and initial emotional features and initial tone features can be further extracted and input into the emotion recognition model and tone recognition model respectively to obtain emotional features and tone features, thus realizing the acquisition of emotional features and tone features in various situations.
[0135] In some optional implementations of this embodiment, the initial generation module 305 can also be used to: concatenate the call vector, emotional features and tone features corresponding to the customer service call in a preset order to obtain the initial composite vector of the customer service call.
[0136] In this embodiment, the call vector, emotional features, and tone features corresponding to the customer service call are concatenated in a preset order to accurately and orderly obtain the initial composite vector of each customer service call.
[0137] In some optional implementations of this embodiment, the weight addition module 306 may include: a score calculation submodule, an average calculation submodule, a weight acquisition submodule, and a weight addition submodule, wherein:
[0138] The score calculation submodule is used to calculate the attention score between each customer service call and other customer service calls.
[0139] The average calculation submodule is used to calculate the average attention score of a customer service call based on the attention score between the customer service call and other customer service calls.
[0140] The weight acquisition submodule is used to normalize the average attention score of each customer service call to obtain the attention weight of each customer service call.
[0141] The weighting submodule is used to add the attention weights of each customer service call to the initial composite vector of each customer service call, thus obtaining the composite vector of each customer service call.
[0142] In this embodiment, for each customer service call, the attention score between the customer service call and other customer service calls is calculated. Based on the attention scores between the customer service call and other customer service calls, the average attention score of the customer service call is calculated. The attention score reflects the similarity or importance of the customer service call to other customer service calls, while the average attention score reflects the average similarity or correlation between the customer service call and other customer service calls. The stronger the correlation, the more attention should be paid. After normalizing the average attention score, the resulting attention weight is also higher. The attention weight of each customer service call is added to the initial composite vector of each customer service call to obtain the composite vector of each customer service call. Adding the attention weight to the vector ensures that the subsequent customer service call quality inspection can be more accurate.
[0143] In some optional implementations of this embodiment, the score calculation submodule can also be used to: for each customer service call, calculate the similarity between the call vector of the customer service call and the call vectors of other customer service calls, and use the obtained similarity as the attention score between the customer service call and other customer service calls.
[0144] In this embodiment, the similarity between the call vector of a customer service call and the call vectors of other customer service calls is calculated, and the obtained similarity is used as the attention score between the customer service call and other customer service calls, thus realizing the calculation of attention score and the introduction of attention.
[0145] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.
[0146] 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, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.
[0147] 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.
[0148] 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 call 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.
[0149] 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 call quality inspection method.
[0150] 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.
[0151] The computer device provided in this embodiment can execute the above-described customer service call quality inspection method. The customer service call quality inspection method here can be any of the methods described in the various embodiments above.
[0152] In this embodiment, customer service call information is obtained. For each sentence of the customer service call in the call information, the call is embedded and encoded using a large-scale natural language pre-trained model to more accurately generate a call vector. A type recognition model is used to perform quality inspection type identification on the call vectors of each customer service call to determine the quality inspection type involved in the call information. To obtain information beyond the literal meaning, an emotion recognition model is used to obtain the emotional features of the customer service call to capture the emotional tendency in the call. A tone recognition model is used to obtain the tone features of the customer service call to analyze the tone in the call. Based on the call vector, emotion features, and tone features corresponding to the customer service call, an initial composite vector of the customer service call is generated, integrating multiple... The application incorporates multi-dimensional features to improve the accuracy of quality inspection. Attention weights are added to the initial composite vectors of each customer service call to obtain composite vectors for each call, allowing the model to focus more on important calls, further enhancing accuracy. The composite vectors of each call are input into the detection models for each quality inspection point under each quality inspection type. Each detection model performs binary classification prediction for a specific quality inspection point, accurately outputting the detection results at that point. Based on these results, the call quality inspection results are generated, further improving the accuracy of the call quality inspection results. This application combines multi-dimensional features with a multi-class hierarchical model to progressively inspect call information, significantly improving the accuracy and efficiency of customer service call quality inspection.
[0153] 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 call quality inspection method described above.
[0154] In this embodiment, customer service call information is obtained. For each sentence of the customer service call in the call information, the call is embedded and encoded using a large-scale natural language pre-trained model to more accurately generate a call vector. A type recognition model is used to perform quality inspection type identification on the call vectors of each customer service call to determine the quality inspection type involved in the call information. To obtain information beyond the literal meaning, an emotion recognition model is used to obtain the emotional features of the customer service call to capture the emotional tendency in the call. A tone recognition model is used to obtain the tone features of the customer service call to analyze the tone in the call. Based on the call vector, emotion features, and tone features corresponding to the customer service call, an initial composite vector of the customer service call is generated, integrating multiple... The application incorporates multi-dimensional features to improve the accuracy of quality inspection. Attention weights are added to the initial composite vectors of each customer service call to obtain composite vectors for each call, allowing the model to focus more on important calls, further enhancing accuracy. The composite vectors of each call are input into the detection models for each quality inspection point under each quality inspection type. Each detection model performs binary classification prediction for a specific quality inspection point, accurately outputting the detection results at that point. Based on these results, the call quality inspection results are generated, further improving the accuracy of the call quality inspection results. This application combines multi-dimensional features with a multi-class hierarchical model to progressively inspect call information, significantly improving the accuracy and efficiency of customer service call quality inspection.
[0155] 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.
[0156] 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 customer service calls, characterized in that, Includes the following steps: Obtain customer service call information; For each customer service call in the call information, the customer service call is embedded and encoded using a pre-trained model to obtain the call vector of the customer service call; The call vectors of each customer service call are identified using a type recognition model to determine the quality inspection type involved in the call information. The emotional features of the customer service call are obtained through an emotion recognition model, and the tone features of the customer service call are obtained through a tone recognition model. Based on the call vector, emotional features, and tone features corresponding to the customer service call, an initial composite vector for the customer service call is generated. Attention weights are added to the initial composite vectors of each customer service call to obtain the composite vectors of each customer service call; The composite vector of each customer service call is input into the quality inspection point detection model under the quality inspection type to obtain the point detection results output by each quality inspection point detection model, and the call quality inspection result of the customer service is generated based on the point detection results.
2. The customer service call quality inspection method according to claim 1, characterized in that, Before the step of obtaining customer service call information, the following is also included: Obtain a customer service call quality inspection dataset, which includes quality inspection standard information and quality inspection sample information; The initial pre-trained model was fine-tuned based on the customer service call quality inspection dataset to obtain the pre-trained model.
3. The customer service call quality inspection method according to claim 1, characterized in that, The steps for obtaining customer service call information include: When a customer service representative is detected to be in a communication state, real-time call information is obtained from that representative via Kafka; or, Based on the triggered quality inspection instruction, retrieve historical call information from the database.
4. The customer service call quality inspection method according to claim 1, characterized in that, The steps of obtaining the emotional features of the customer service call through an emotion recognition model and obtaining the tone features of the customer service call through a tone recognition model include: The customer service call in text form is input into an emotion recognition model to obtain the emotion features of the customer service call, and the customer service call in text form is input into a tone recognition model to obtain the tone features of the customer service call; or, The acoustic features of the customer service call in speech form are extracted, and initial emotional features and initial tone features are extracted from the acoustic features; The initial emotional features are input into the emotional recognition model to obtain the emotional features of the customer service call, and the initial tone features are input into the tone recognition model to obtain the tone features of the customer service call.
5. The customer service call quality inspection method according to claim 1, characterized in that, The step of generating an initial composite vector for the customer service call based on the call vector, emotional features, and tone features corresponding to the customer service call includes: The call vector, emotional features, and tone features corresponding to the customer service call are concatenated in a preset order to obtain the initial composite vector of the customer service call.
6. The customer service call quality inspection method according to claim 1, characterized in that, The step of adding attention weights to the initial composite vectors of each customer service call to obtain the composite vectors of each customer service call includes: For each customer service call, calculate the attention score between the customer service call and all other customer service calls; Calculate the average attention score of the customer service call based on the attention scores between the customer service call and each of the other customer service calls; The average attention score of each customer service call is normalized to obtain the attention weight of each customer service call; The attention weights of each customer service call are added to the initial composite vector of each customer service call to obtain the composite vector of each customer service call.
7. The customer service call quality inspection method according to claim 6, characterized in that, The step of calculating the attention score between each customer service call and all other customer service calls: For each customer service call, calculate the similarity between the call vector of the customer service call and the call vectors of all other customer service calls, and use the obtained similarity as the attention score between the customer service call and the other customer service calls.
8. A customer service call quality inspection device, characterized in that, include: The call acquisition module is used to acquire call information from customer service representatives. The call encoding module is used to embed and encode each customer service call in the call information using a pre-trained model to obtain the call vector of the customer service call. The type recognition module is used to identify the quality inspection type of each customer service call based on the type recognition model, and obtain the quality inspection type involved in the call information. The feature acquisition module is used to acquire the emotional features of the customer service call through an emotion recognition model and to the tone features of the customer service call through a tone recognition model. The initial generation module is used to generate an initial composite vector for the customer service call based on the call vector, emotional features, and tone features corresponding to the customer service call. The weighting module is used to add attention weights to the initial composite vector of each customer service call to obtain the composite vector of each customer service call. The call quality inspection module is used to input the composite vector of each customer service call into the quality inspection point detection model under the quality inspection type, obtain the point detection results output by each quality inspection point detection model, and generate the call quality inspection result of the customer service based on the point detection results.
9. 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 call quality inspection method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the customer service call quality inspection method as described in any one of claims 1 to 7.