Quality detection method applied to seat call and related equipment
Through the combination of pre-trained combination model and processing combination model, the problem of low accuracy of traditional quality inspection methods is solved, efficient agent call quality inspection is achieved, the risk of missed inspection is reduced and compliance risk control accuracy is improved.
Patent Information
- Application Number
- CN202411190812.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-12-06
AI Technical Summary
Traditional artificial quality inspection methods are very labor-intensive and easy to miss inspection. Machine quality inspection systems based on keywords and regular expressions are difficult to deeply explore subtle differences and potential risks in call content, resulting in low accuracy of quality inspection results.
Pre-trained combinatorial models (such as RoBERTa, SpanBERT, KBERT, ALBERT, ELECTRA) are used for language understanding and text pre-training, and classify and capture processing of processing combinatorial models (such as TextCNN, TextRNN, Bi-LSTM) and score through the full connection layer. Predict, output quality inspection results.
Significantly reduce the workload of artificial quality inspection, reduce the risk of missed inspection, improve the accuracy and efficiency of compliance risk control, can better adapt to diversified business scenarios, and help enterprises discover potential business opportunities and public opinion risks.
Smart Images

Figure CN119094658A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a quality detection method and related equipment for agent calls. Background Art
[0002] With the expansion of the company's business and the intensification of market competition, outbound calling has become a key channel for enterprises to communicate with customers. In this context, standardized operation of seats and effective monitoring of user public opinion risks are particularly important. In order to ensure service quality and compliance, quality inspection operations have become an indispensable link.
[0003] However, the applicant found that although the traditional manual quality inspection method is intuitive, it faces challenges such as large workload, easy to miss detection, and difficulty in discovering potential problems. In addition, with the diversification of business scenarios, traditional machine quality inspection systems based on keywords and regular expressions have been unable to meet increasingly stringent compliance and risk control requirements. Although these systems are easy to deploy, they have limited effects in complex and changeable practical applications and cannot deeply explore the subtle differences and potential risks in the content of calls. It can be seen that the traditional quality inspection method has the problem of low accuracy of quality inspection results. Summary of the invention
[0004] The purpose of the embodiments of the present application is to provide a quality detection method and related equipment for agent calls to solve the problem of low accuracy of quality detection results in traditional quality detection methods.
[0005] In order to solve the above technical problems, the embodiment of the present application provides a quality detection method for agent calls, which adopts the following technical solution:
[0006] Get the call data of the agent to be tested;
[0007] Performing a pre-training operation on the agent call data according to the pre-training combination model to obtain a pre-training text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model;
[0008] Performing classification capture processing on the pre-trained text according to the processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model, and a Bi-LSTM model;
[0009] Inputting the text classification result into the fully connected layer to perform a score prediction operation to obtain a score prediction result;
[0010] Output the score prediction result.
[0011] Furthermore, the step of performing a pre-training operation on the agent call data according to the pre-trained combined model to obtain a pre-trained text specifically includes the following steps:
[0012] Performing language understanding analysis on the agent call data according to the RoBERTa model to obtain a first semantic text;
[0013] Performing a long-distance dependency analysis on the first semantic text according to the SpanBERT model to obtain a second semantic text;
[0014] Recognize and interpret the second semantic text according to the KBERT model to obtain a third semantic text;
[0015] Optimizing according to the ALBERT model and the third semantic text to obtain a fourth semantic text generated by the model after parameter optimization;
[0016] Correcting the fourth semantic text according to the ELECTRA model to obtain a fifth semantic text;
[0017] A fusion operation is performed on the first semantic text, the second semantic text, the third semantic text, the fourth semantic text, and the fifth semantic text to obtain the pre-trained text.
[0018] Furthermore, the step of performing classification capture processing on the pre-trained text according to the processing combination model to obtain a text classification result specifically includes the following steps:
[0019] Extract local features of the pre-trained text according to the TextCNN model to obtain local feature data;
[0020] Extracting sequence features from the pre-trained text according to the TextRNN model to obtain sequence feature data;
[0021] Performing bidirectional processing on the pre-trained text according to the Bi-LSTM model to obtain bidirectional context data;
[0022] fusing the local feature data, the sequence feature data and the bidirectional context data to obtain fused feature data;
[0023] The fused feature data is subjected to category prediction according to the softmax function to obtain the text classification result.
[0024] Furthermore, after the step of performing classification capture processing on the pre-trained text according to the processing combination model to obtain a text classification result, the following steps are also included:
[0025] Compare the text classification result with the preset standard result to obtain a classification comparison result;
[0026] According to the classification comparison result, the convolution layer parameters of the TextCNN model are adjusted, the weight of the TextRNN model is updated, and the Bi-LSTM model is repeatedly trained to obtain an optimized processing combination model.
[0027] Furthermore, the step of obtaining the agent call data to be detected specifically includes the following steps:
[0028] Obtain the call recording files between the agent and the customer through the voice interface;
[0029] The call recording file is transcribed according to the speech recognition technology to obtain the voice call text;
[0030] Performing object recognition on the voice call text according to natural language processing technology to obtain a voice call text marked with a call object identifier;
[0031] The emotion analysis is performed on the voice call text marked with the call object identifier according to the emotion recognition model to obtain the emotion label of each voice call.
[0032] In order to solve the above technical problems, the embodiment of the present application further provides a quality detection device for agent calls, which adopts the following technical solution:
[0033] A call data acquisition module is used to acquire the call data of the agent to be detected;
[0034] A pre-training module, used for performing a pre-training operation on the agent call data according to a pre-training combination model to obtain a pre-training text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model;
[0035] A classification capture module, used for performing classification capture processing on the pre-trained text according to a processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model and a Bi-LSTM model;
[0036] A score prediction module, used to input the text classification result into the fully connected layer to perform a score prediction operation to obtain a score prediction result;
[0037] The result output module is used to output the score prediction result.
[0038] Furthermore, the pre-training module includes:
[0039] A language understanding and analysis submodule, configured to perform language understanding and analysis on the agent call data according to the RoBERTa model to obtain a first semantic text;
[0040] A dependency analysis submodule, configured to perform a long-distance dependency analysis on the first semantic text according to the SpanBERT model to obtain a second semantic text;
[0041] A recognition and interpretation submodule, used for recognizing and interpreting the second semantic text according to the KBERT model to obtain a third semantic text;
[0042] A parameter optimization submodule, used for performing optimization according to the ALBERT model and the third semantic text to obtain a fourth semantic text generated by the model after parameter optimization;
[0043] A text correction submodule, used for correcting the fourth semantic text according to the ELECTRA model to obtain a fifth semantic text;
[0044] The text fusion submodule is used to perform a fusion operation on the first semantic text, the second semantic text, the third semantic text, the fourth semantic text and the fifth semantic text to obtain the pre-trained text.
[0045] Furthermore, the classification capture module includes:
[0046] A local feature extraction submodule, used to extract local features of the pre-trained text according to the TextCNN model to obtain local feature data;
[0047] A sequence feature extraction submodule, used to extract sequence features from the pre-trained text according to the TextRNN model to obtain sequence feature data;
[0048] A bidirectional processing submodule, used for performing bidirectional processing on the pre-trained text according to the Bi-LSTM model to obtain bidirectional context data;
[0049] A feature fusion submodule, used for fusing the local feature data, the sequence feature data and the bidirectional context data to obtain fused feature data;
[0050] The category prediction submodule is used to perform category prediction on the fused feature data according to the softmax function to obtain the text classification result.
[0051] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:
[0052] The system comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the quality detection method applied to agent calls as described above when executing the computer-readable instructions.
[0053] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:
[0054] The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the quality detection method applied to the agent call as described above are implemented.
[0055] The present application provides a quality detection method for agent calls, including: obtaining agent call data to be detected; performing pre-training operations on the agent call data according to a pre-training combination model to obtain pre-trained text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model; performing classification and capture processing on the pre-trained text according to a processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model, and a Bi-LSTM model; inputting the text classification result into a fully connected layer for score prediction operations to obtain a score prediction result; and outputting the score prediction result. Compared with the prior art, the present application can significantly reduce the workload of manual quality inspection and reduce the risk of missed inspections due to manual operations through its highly automated intelligent processing process. At the same time, compared with traditional machine quality inspection systems based on keywords and regular expressions, the present invention can better adapt to diversified business scenarios and improve the accuracy and efficiency of compliance risk control. In addition, the present invention can also help enterprises discover potential business opportunities and public opinion risks, thereby playing an important role in maintaining corporate image and improving customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0057] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;
[0058] Figure 2 is a flow chart of the implementation of the quality detection method for agent calls provided in the first embodiment of the present application;
[0059] Figure 3 It is a structural diagram of a quality detection device for agent calls provided in Embodiment 2 of the present application;
[0060] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION
[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.
[0062] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0063] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.
[0064] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.
[0065] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.
[0066] The terminal devices 101, 102, and 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, etc.
[0067] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .
[0068] It should be noted that the quality detection method for seat calls provided in the embodiments of the present application is generally executed by a server / terminal device, and accordingly, the quality detection device for seat calls is generally arranged in a server / terminal device.
[0069] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.
[0070] Embodiment 1
[0071] Continue to refer Figure 2 , shows a flow chart of an embodiment of a quality detection method for seat calls according to the present application. The quality detection method for seat calls includes: step S201, step S202, step S203, step S204, step S205, step S206 and step S207.
[0072] In step S201, the call data of the agent to be detected is obtained.
[0073] In an embodiment of the present application, the original call data can be extracted by accessing the call record database to obtain the agent call data and provide initial text material for the subsequent pre-training model.
[0074] In an embodiment of the present application, after acquiring the agent call data, a data cleaning script written in Python can be used to remove background noise and non-language components in all call recordings, and precise cleaning can be achieved using regular expressions and custom filtering rules.
[0075] In an embodiment of the present application, the ji eba word segmentation tool can be used to perform word segmentation on the cleaned text to cut continuous text strings into separate vocabulary units.
[0076] In step S202, a pre-training operation is performed on the agent call data according to the pre-training combination model to obtain a pre-training text, wherein the pre-training combination model consists of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model.
[0077] In an embodiment of the present application, during the model training stage, models such as RoBERTa, SpanBERT, KBERT, ALBERT and ELECTRA can be integrated through a specific framework. Specifically, the TensorFlow or PyTorch framework can be used, and appropriate parameters such as learning rate 001, batch size 32, etc. can be set to initialize the combined model.
[0078] In the embodiment of the present application, these models are pre-trained on the training set, and through multiple rounds of iterations (such as 50 rounds), the internal parameters are continuously adjusted to adapt to the specific language environment of the agent call.
[0079] In an embodiment of the present application, during the performance evaluation phase, an independent validation dataset is used to evaluate the performance of each model component in detail by calculating indicators such as accuracy, recall, and F1 score.
[0080] In an embodiment of the present application, according to the evaluation results, the hyperparameters or model structure are adjusted, such as modifying the number of hidden layers of RoBERTa or adjusting the number of attention heads of SpanBERT, to improve the overall performance.
[0081] In the embodiment of the present application, the tuned model is applied to the full amount of agent call data to generate pre-trained texts. These texts not only contain high-quality language features, but also adapt to specific call scenarios, thus providing a strong support foundation for downstream NLP tasks.
[0082] In the embodiments of the present application, the pre-training effect of the model can be effectively improved, thereby achieving higher accuracy and efficiency in practical applications.
[0083] In step S203, the pre-trained text is classified and captured according to the processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model and a Bi-LSTM model.
[0084] In the embodiment of the present application, when the text is classified, the pre-trained text data is first processed by the TextCNN model, and the number of filters of the convolutional layer is set to 256, and the size of each filter is 3x3, so as to capture the local features in the text and obtain the local semantic representation. These local semantic representations are then input into the TextRNN model, which adopts a recurrent neural network structure with 256 hidden units, effectively captures the long-distance dependencies of the text, and integrates the information into a text representation containing time series features. In order to further refine the text information, the representation is processed by a Bi-LSTM model, which is configured with a bidirectional long short-term memory network, and 128 memory units are set in each direction, which strengthens the comprehensive understanding of the forward and backward information of the text, thereby generating a richer text context representation. Next, based on the text context representation output by the Bi-LSTM model, a multi-layer perceptron (MLP) containing two hidden layers is used for deep semantic analysis, each hidden layer is provided with 1024 neurons, and ReLU is used as the activation function, so as to perform advanced classification processing. In this process, the MLP model classifies the text into specific categories based on the comprehensive semantic representation, and finally generates the classification results of the text.
[0085] In step S204, the text classification result is input into the fully connected layer for score prediction operation to obtain a score prediction result.
[0086] In an embodiment of the present application, after obtaining the text classification result, the text classification result is passed as an input feature to a fully connected neural network, which predicts the quality inspection score of each call based on the input features, wherein the prediction process includes a multi-layer neural network and an activation function, thereby ensuring the accuracy and reliability of the output result.
[0087] In step S205, the score prediction result is output.
[0088] In an embodiment of the present application, a quality detection method for agent calls is provided, including: obtaining agent call data to be detected; performing pre-training operations on the agent call data according to a pre-training combination model to obtain pre-trained text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model; performing classification and capture processing on the pre-trained text according to the processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model, and a Bi-LSTM model; inputting the text classification result into a fully connected layer for score prediction operations to obtain a score prediction result; and outputting the score prediction result. Compared with the prior art, the present application can significantly reduce the workload of manual quality inspection and reduce the risk of missed inspections caused by manual operations through its highly automated intelligent processing process. At the same time, compared with the traditional machine quality inspection system based on keywords and regular expressions, the present invention can better adapt to diversified business scenarios and improve the accuracy and efficiency of compliance risk control. In addition, the present invention can also help enterprises discover potential business opportunities and public opinion risks, thereby playing an important role in maintaining corporate image and improving customer satisfaction.
[0089] In some optional implementations of the embodiments of the present application, step S202 includes the following steps:
[0090] Perform language understanding analysis on the agent call data according to the RoBERTa model to obtain the first semantic text;
[0091] Perform a long-distance dependency analysis on the first semantic text according to the SpanBERT model to obtain a second semantic text;
[0092] The second semantic text is recognized and interpreted according to the KBERT model to obtain a third semantic text;
[0093] Optimizing the ALBERT model and the third semantic text to obtain a fourth semantic text generated by the model after parameter optimization;
[0094] The fourth semantic text is corrected according to the ELECTRA model to obtain the fifth semantic text;
[0095] A fusion operation is performed on the first semantic text, the second semantic text, the third semantic text, the fourth semantic text, and the fifth semantic text to obtain a pre-trained text.
[0096] In the embodiment of the present application, the RoBERTa model is used to perform basic semantic understanding of the discourse, such as identifying topics and intentions. At this stage, the model is fine-tuned by using the transformers library, adjusting the learning rate to 5e-5, and setting the batch size to 32 to adapt to the specific data set.
[0097] In an embodiment of the present application, the SpanBERT model is used to enhance the understanding of long-distance semantic dependencies. The SpanBERT model pays special attention to the relationship between sentences in the conversation, how to capture information through links across large sections of text, and uses a self-attention mechanism to improve the ability to capture context.
[0098] In an embodiment of the present application, an external knowledge base is introduced through the KBERT model, for example, a knowledge graph in the medical or legal field is used, and the professional terms appearing in the call content are matched with the entities in the knowledge base through entity linking technology. Then, a graph neural network is used to perform in-depth analysis on these connected entities, thereby enhancing the model's depth of understanding of professional content.
[0099] In the embodiments of the present application, the ALBERT model is used to optimize the above model parameters, especially to improve the model size and operating efficiency. By reducing the number of layers and hidden units of the model, the model is made lighter while maintaining performance to achieve fast data processing.
[0100] In the embodiment of the present application, the ELECTRA model is used to detect and automatically correct errors in the integrated text to ensure the accuracy and robustness of the text. The model uses an adversarial generative network to identify and correct subtle errors in the generated text, further improving the quality of the text.
[0101] In an embodiment of the present application, after integrating the output results of all models, the data is fused to form a complete pre-trained text.
[0102] In some optional implementations of the embodiments of the present application, after step S203, the following steps are further included:
[0103] Compare the text classification result with the preset standard result to obtain the classification comparison result;
[0104] According to the classification comparison results, the convolutional layer parameters of the TextCNN model are adjusted, the weights of the TextRNN model are updated, and the Bi-LSTM model is repeatedly trained to obtain the optimized processing combination model.
[0105] In an embodiment of the present application, in order to improve the accuracy of the evaluation classification, the above text classification results are compared with the preset standards, and the classification accuracy and confusion matrix are calculated to generate the classification comparison results. According to the feedback of the classification comparison results, the convolution layer parameters in the TextCNN model are adjusted, the filter size is optimized to 4x4, and the number of filters is increased to 320 to improve the ability to capture local features; at the same time, the weights of the TextRNN model are updated, and the number of hidden units in the loop layer is adjusted to 320 to enhance the ability to capture long-distance dependencies; in addition, the Bi-LSTM model has also been retrained, and the number of memory units in each direction of the adjusted model in the bidirectional network has been increased to 160, further enhancing the forward and backward processing capabilities of text information.
[0106] In the embodiment of the present application, through the above parameter optimization and model training, the effect of the entire text processing process has been significantly improved, providing more accurate and comprehensive support for the final text classification. It not only improves the performance of the model, but also provides a solid foundation for subsequent text analysis and application.
[0107] In some optional implementations of the embodiments of the present application, step S203 specifically includes the following steps:
[0108] Extract local features of the pre-trained text according to the TextCNN model to obtain local feature data;
[0109] Extract sequence features from pre-trained text according to the TextRNN model to obtain sequence feature data;
[0110] The pre-trained text is processed bidirectionally according to the Bi-LSTM model to obtain bidirectional context data;
[0111] Fusing local feature data, sequence feature data and bidirectional context data to obtain fused feature data;
[0112] The category of the fused feature data is predicted according to the softmax function to obtain the text classification result.
[0113] In an embodiment of the present application, a TextCNN model is used to extract local features of the text, and multiple convolution kernels are constructed. The sizes are set to 3, 4, and 5, respectively. The number of convolution kernels of each size is 128, which is used to capture word combinations of different lengths, thereby identifying key phrases or vocabulary.
[0114] In the embodiment of the present application, the TextRNN model is combined, especially the LSTM unit is used to process sequence dependency, and 128 hidden layer units are set to capture long-distance dependencies to enhance the understanding of contextual coherence.
[0115] In the embodiment of the present application, in order to fully grasp the text information, the Bi-LSTM model is further introduced to perform bidirectional analysis of the text. 64 LSTM units are set in each direction, which can simultaneously obtain and integrate the forward and backward semantic information of the text, which is important for understanding the meaning of the text.
[0116] In an embodiment of the present application, by fusing the outputs of TextCNN, TextRNN and Bi-LSTM, a connection operation is used to form a comprehensive feature vector, which integrates information from different models and expresses the text features more comprehensively.
[0117] In the embodiment of the present application, the fused feature vector is input to the fully connected layer for classification processing. The fully connected layer has 256 neurons and uses the ReLU activation function to increase the nonlinear processing capability. The output layer uses the softmax function to convert the feature vector into a probability distribution of each category, so that the call quality can be judged according to the category with the highest probability, realizing the effective identification of service quality issues.
[0118] In the embodiments of the present application, service quality-related issues can be accurately identified and classified from a large amount of call data, providing scientific data support for subsequent service quality improvements.
[0119] In some optional implementations of the embodiments of the present application, step S201 specifically includes the following steps:
[0120] Obtain the call recording files between the agent and the customer through the voice interface;
[0121] The call recording file is transcribed according to the speech recognition technology to obtain the voice call text;
[0122] Performing object recognition on the voice call text according to natural language processing technology to obtain the voice call text marked with the call object identifier;
[0123] The emotion recognition model is used to perform emotion analysis on the voice call text marked with the caller ID to obtain the emotion label of each voice call.
[0124] The embodiments of the present 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 knowledge to obtain the best results.
[0125] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0126] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the 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 above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).
[0127] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.
[0128] Embodiment 2
[0129] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a quality detection device for seat call, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.
[0130] like Figure 3 As shown, the quality detection device 200 applied to the seat call in the embodiment of the present application includes:
[0131] The call data acquisition module 210 is used to acquire the call data of the seat to be detected;
[0132] A pre-training module 220 is used to perform a pre-training operation on the agent call data according to the pre-training combination model to obtain a pre-training text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model;
[0133] A classification capture module 230 is used to perform classification capture processing on the pre-trained text according to the processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model and a Bi-LSTM model;
[0134] The score prediction module 240 is used to input the text classification result into the fully connected layer to perform a score prediction operation to obtain a score prediction result;
[0135] The result output module 250 is used to output the score prediction result.
[0136] In an embodiment of the present application, a quality detection device 200 for agent calls is provided, including: a call data acquisition module 210, used to obtain agent call data to be detected; a pre-training module 220, used to perform a pre-training operation on the agent call data according to a pre-training combination model to obtain a pre-trained text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model and an ELECTRA model; a classification capture module 230, used to perform a classification capture process on the pre-training text according to a processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model and a Bi-LSTM model; a score prediction module 240, used to input the text classification result into a fully connected layer for a score prediction operation to obtain a score prediction result; a result output module 250, used to output the score prediction result. Compared with the prior art, this application can significantly reduce the workload of manual quality inspection and the risk of missed inspections due to manual operations through its highly automated intelligent processing flow. At the same time, compared with the traditional machine quality inspection system based on keywords and regular expressions, the present invention can better adapt to diversified business scenarios and improve the accuracy and efficiency of compliance risk control. In addition, the present invention can also help enterprises discover potential business opportunities and public opinion risks, thereby playing an important role in maintaining corporate image and improving customer satisfaction.
[0137] In some optional implementations of the embodiments of the present application, the pre-training module includes:
[0138] A language understanding and analysis submodule is used to perform language understanding and analysis on the agent call data according to the RoBERTa model to obtain a first semantic text;
[0139] A dependency analysis submodule, used for performing a long-distance dependency analysis on the first semantic text according to the SpanBERT model to obtain a second semantic text;
[0140] The recognition and interpretation submodule is used to recognize and interpret the second semantic text according to the KBERT model to obtain the third semantic text;
[0141] A parameter optimization submodule, used for performing optimization according to the ALBERT model and the third semantic text to obtain a fourth semantic text generated by the model after parameter optimization;
[0142] A text correction submodule, used for correcting the fourth semantic text according to the ELECTRA model to obtain a fifth semantic text;
[0143] The text fusion submodule is used to perform a fusion operation on the first semantic text, the second semantic text, the third semantic text, the fourth semantic text and the fifth semantic text to obtain a pre-trained text.
[0144] In some optional implementations of the embodiments of the present application, the classification capture module includes:
[0145] The local feature extraction submodule is used to extract local features of the pre-trained text according to the TextCNN model to obtain local feature data;
[0146] The sequence feature extraction submodule is used to extract sequence features from the pre-trained text according to the TextRNN model to obtain sequence feature data;
[0147] The bidirectional processing submodule is used to perform bidirectional processing on the pre-trained text according to the Bi-LSTM model to obtain bidirectional context data;
[0148] The feature fusion submodule is used to fuse local feature data, sequence feature data and bidirectional context data to obtain fused feature data;
[0149] The category prediction submodule is used to perform category prediction on the fused feature data according to the softmax function to obtain the text classification result.
[0150] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of a computer device according to an embodiment of the present application.
[0151] The computer device 300 includes a memory 310, a processor 320, and a network interface 330 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 300 having components 310-330, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform 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 processors (DSPs), embedded devices, etc.
[0152] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.
[0153] The memory 310 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), 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 310 can be an internal storage unit of the computer device 300, such as a hard disk or memory of the computer device 300. In other embodiments, the memory 310 can also be an external storage device of the computer device 300, such as a plug-in hard disk equipped on the computer device 300, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Of course, the memory 310 can also include both the internal storage unit of the computer device 300 and its external storage device. In the embodiment of the present application, the memory 310 is generally used to store the operating system and various application software installed on the computer device 300, such as computer-readable instructions for the quality detection method of the agent call, etc. In addition, the memory 310 can also be used to temporarily store various data that have been output or are to be output.
[0154] The processor 320 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 320 is generally used to control the overall operation of the computer device 300. In the embodiment of the present application, the processor 320 is used to run the computer-readable instructions stored in the memory 310 or process data, such as running the computer-readable instructions of the quality detection method applied to the agent call.
[0155] The network interface 330 may include a wireless network interface or a wired network interface. The network interface 330 is generally used to establish a communication connection between the computer device 300 and other electronic devices.
[0156] The computer equipment provided by this application can significantly reduce the workload of manual quality inspection and the risk of missed inspections due to manual operations through its highly automated intelligent processing flow. At the same time, compared with the traditional machine quality inspection system based on keywords and regular expressions, the present invention can better adapt to diversified business scenarios and improve the accuracy and efficiency of compliance risk control. In addition, the present invention can also help enterprises discover potential business opportunities and public opinion risks, thereby playing an important role in maintaining corporate image and improving customer satisfaction.
[0157] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the quality detection method applied to agent calls as described above.
[0158] The computer-readable storage medium provided by the present application can significantly reduce the workload of manual quality inspection and the risk of missed inspections due to manual operations through its highly automated intelligent processing flow. At the same time, compared with the traditional machine quality inspection system based on keywords and regular expressions, the present invention can better adapt to diversified business scenarios and improve the accuracy and efficiency of compliance risk control. In addition, the present invention can also help enterprises discover potential business opportunities and public opinion risks, thereby playing an important role in maintaining corporate image and improving customer satisfaction.
[0159] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.
[0160] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.
Claims
1. A quality detection method for agent calls, characterized in that: The steps include: Get the call data of the agent to be tested; Performing a pre-training operation on the agent call data according to the pre-training combination model to obtain a pre-training text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model; Performing classification capture processing on the pre-trained text according to the processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model, and a Bi-LSTM model; Inputting the text classification result into the fully connected layer to perform a score prediction operation to obtain a score prediction result; Output the score prediction result.
2. The quality detection method for seat calls according to claim 1 is characterized in that: The step of performing a pre-training operation on the agent call data according to the pre-training combination model to obtain a pre-trained text specifically includes the following steps: Performing language understanding analysis on the agent call data according to the RoBERTa model to obtain a first semantic text; Performing a long-distance dependency analysis on the first semantic text according to the SpanBERT model to obtain a second semantic text; Recognize and interpret the second semantic text according to the KBERT model to obtain a third semantic text; Optimizing according to the ALBERT model and the third semantic text to obtain a fourth semantic text generated by the model after parameter optimization; Correcting the fourth semantic text according to the ELECTRA model to obtain a fifth semantic text; A fusion operation is performed on the first semantic text, the second semantic text, the third semantic text, the fourth semantic text, and the fifth semantic text to obtain the pre-trained text.
3. The quality detection method for seat calls according to claim 1 is characterized in that: The step of performing classification capture processing on the pre-trained text according to the processing combination model to obtain a text classification result specifically includes the following steps: Extract local features of the pre-trained text according to the TextCNN model to obtain local feature data; Extracting sequence features from the pre-trained text according to the TextRNN model to obtain sequence feature data; Performing bidirectional processing on the pre-trained text according to the Bi-LSTM model to obtain bidirectional context data; fusing the local feature data, the sequence feature data and the bidirectional context data to obtain fused feature data; The fused feature data is subjected to category prediction according to the softmax function to obtain the text classification result.
4. The quality detection method for seat calls according to claim 1, characterized in that: After the step of performing classification and capture processing on the pre-trained text according to the processing combination model to obtain a text classification result, the following steps are also included: Compare the text classification result with the preset standard result to obtain a classification comparison result; According to the classification comparison result, the convolution layer parameters of the TextCNN model are adjusted, the weight of the TextRNN model is updated, and the Bi-LSTM model is repeatedly trained to obtain an optimized processing combination model.
5. The quality detection method for seat calls according to claim 1 is characterized in that: The step of obtaining the agent call data to be detected specifically includes the following steps: Obtain the call recording files between the agent and the customer through the voice interface; The call recording file is transcribed according to the speech recognition technology to obtain the voice call text; Performing object recognition on the voice call text according to natural language processing technology to obtain a voice call text marked with a call object identifier; The emotion analysis is performed on the voice call text marked with the call object identifier according to the emotion recognition model to obtain the emotion label of each voice call.
6. A quality detection device for seat calls, characterized in that: include: A call data acquisition module is used to acquire the call data of the agent to be detected; A pre-training module, used for performing a pre-training operation on the agent call data according to a pre-training combination model to obtain a pre-training text, wherein the pre-training combination model is composed of a RoBERTa model, a SpanBERT model, a KBERT model, an ALBERT model, and an ELECTRA model; A classification capture module, used for performing classification capture processing on the pre-trained text according to a processing combination model to obtain a text classification result, wherein the processing combination model is composed of a TextCNN model, a TextRNN model and a Bi-LSTM model; A score prediction module, used to input the text classification result into the fully connected layer to perform a score prediction operation to obtain a score prediction result; The result output module is used to output the score prediction result.
7. The quality detection device for seat calls according to claim 6, characterized in that: The pre-training module includes: A language understanding and analysis submodule, configured to perform language understanding and analysis on the agent call data according to the RoBERTa model to obtain a first semantic text; A dependency analysis submodule, configured to perform a long-distance dependency analysis on the first semantic text according to the SpanBERT model to obtain a second semantic text; A recognition and interpretation submodule, used for recognizing and interpreting the second semantic text according to the KBERT model to obtain a third semantic text; A parameter optimization submodule, used for performing optimization according to the ALBERT model and the third semantic text to obtain a fourth semantic text generated by the model after parameter optimization; A text correction submodule, used for correcting the fourth semantic text according to the ELECTRA model to obtain a fifth semantic text; The text fusion submodule is used to perform a fusion operation on the first semantic text, the second semantic text, the third semantic text, the fourth semantic text and the fifth semantic text to obtain the pre-trained text.
8. The quality detection device for seat calls according to claim 6, characterized in that: The classification capture module comprises: A local feature extraction submodule, used to extract local features of the pre-trained text according to the TextCNN model to obtain local feature data; A sequence feature extraction submodule, used to extract sequence features from the pre-trained text according to the TextRNN model to obtain sequence feature data; A bidirectional processing submodule, used for performing bidirectional processing on the pre-trained text according to the Bi-LSTM model to obtain bidirectional context data; A feature fusion submodule, used for fusing the local feature data, the sequence feature data and the bidirectional context data to obtain fused feature data; The category prediction submodule is used to perform category prediction on the fused feature data according to the softmax function to obtain the text classification result.
9. A computer device comprising a memory and a processor, characterized in that: The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the quality detection method for agent calls according to any one of claims 1 to 5 are implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the quality detection method for agent calls according to any one of claims 1 to 5 are implemented.
