Call center automation service optimization method, apparatus, and medium

Abstract

The application provides a call center automation service optimization method, device and medium, and relates to the technical field of computers. The method comprises the following steps: obtaining the call voice of a customer service and a customer, extracting the first semantic information and the first tone information of the customer service and the second semantic information and the second tone information of the customer in the call voice; obtaining the speech score, the business score and the first emotional score of the customer service and the second emotional score of the customer according to the first semantic information and the first tone information of the customer service and the second semantic information and the second tone information of the customer; and positioning the service problem and providing a strategy suggestion to guide the customer service to optimize the call service according to at least one of the speech score, the business score and the first emotional score of the customer service and the second emotional score of the customer. The application distinguishes the semantics and the tone of the customer service and the customer, evaluates the customer service and the emotion of the customer in multiple dimensions, positions the call service problem, provides a targeted guidance strategy suggestion, and optimizes the call service quality of the customer service.

Description

Technical Field

[0001] This application relates at least to the field of computer technology, and in particular to a method, apparatus and medium for optimizing automated services in a call center. Background Technology

[0002] Because the quality of call center customer service personnel varies, some may lack sufficient product knowledge or communication skills. For example, in a telecommunications service operator's call center, customer service personnel need to accurately explain complex package terms and conditions to customers. If a customer service representative does not have a deep understanding of the product, they may not be able to answer customer questions clearly and accurately, leading to misunderstandings about product information.

[0003] Furthermore, the service attitude of different customer service personnel will vary. Some customer service representatives may be patient and enthusiastic, and able to soothe customers' emotions well; while others may show impatience when faced with customer complaints or complex issues, thus causing customer dissatisfaction.

[0004] Therefore, it is necessary to optimize call center customer service to address the aforementioned issues. Summary of the Invention

[0005] To address the aforementioned shortcomings, this application provides a method, apparatus, and medium for optimizing automated services in call centers, thereby solving the following technical problem: how to accurately identify and locate service problems in customer service calls and provide targeted suggestions to optimize services.

[0006] Firstly, this application provides a method for optimizing automated call center services, the method comprising:

[0007] Obtain the voice recordings of conversations between customer service representatives and customers, and extract the primary semantic information and primary tone information of the customer service representatives, as well as the secondary semantic information and secondary tone information of the customers from the voice recordings.

[0008] Based on the customer service representative's first semantic information and first tone information, and the customer's second semantic information and second tone information, obtain the customer service representative's speech score, business score, first emotional score, and the customer's second emotional score;

[0009] Based on at least one of the customer service representative's communication skills score, business skills score, primary emotional score, and secondary emotional score, identify service issues and provide strategic suggestions to guide customer service representatives in optimizing call services.

[0010] Furthermore, the system acquires the voice recordings of conversations between customer service representatives and customers, and extracts the customer service representative's primary semantic information and primary tone information, as well as the customer's secondary semantic information and secondary tone information from the voice recordings. Specifically, this includes:

[0011] During a call between customer service representatives and a customer, the original audio of the call is captured in real time.

[0012] The original call speech was denoised by combining Mel-frequency cepstral coefficients (MFCC) with a subtraction spectrum algorithm to obtain denoised call speech.

[0013] The noise-reduced call voice is input into the trained semantic tone recognition model to extract the customer service's first semantic information, first tone information, and second semantic information and second tone information from the call voice. The trained semantic tone recognition model consists of a trained language model and algorithms for recognizing speech rate and intonation.

[0014] Furthermore, a combination of Mel-frequency cepstral coefficients (MFCC) and a subtraction spectral algorithm is used to denoise the original call speech to obtain denoised call speech, specifically including:

[0015] The Voice Activity Detection (VAD) algorithm is used to determine the speech segments and non-speech segments of the original call audio.

[0016] The power spectrum of the noisy speech segment and the noise power spectrum of the non-speech segment are obtained by using Mel frequency cepstral coefficients (MFCC).

[0017] The subtraction spectrum algorithm is used to subtract the noise power spectrum from the power spectrum of the noisy speech to obtain the power spectrum of the denoised speech and restore it to the denoised call speech.

[0018] Furthermore, the trained semantic tone recognition large model is obtained through the following steps:

[0019] A large language model is used as the base model. Algorithms for recognizing speech rate and intonation are added to the front end of the base model to obtain an initial large semantic tone recognition model.

[0020] A training dataset is formed by acquiring historical denoised call audio and corresponding historical semantic tone information. The historical semantic tone information includes historical first semantic information, historical first tone information, historical second semantic information and historical second tone information. The tone information is a number of levels that evaluate the call audio as good to bad.

[0021] The training dataset is input into the initial semantic tone recognition model for iterative training. This allows the initial semantic tone recognition model to first identify the speech rate and intonation of historical denoised call speech, and then obtain the recognition results based on the historical denoised call speech and its speech rate and intonation, and compare them with historical semantic tone information, until the training results converge to obtain the trained semantic tone recognition model.

[0022] Furthermore, based on the customer service representative's first semantic information and first tone information, and the customer's second semantic information and second tone information, the customer service representative's communication skills score, business skills score, first emotional score, and the customer's second emotional score are obtained, specifically including:

[0023] Obtain preset customer service script rule keywords and prohibited words. Based on the customer service script rule keywords and prohibited words contained in the first semantic information, and combined with the corresponding preset script word weights, calculate the customer service script score. The customer service script rule keywords and prohibited words are set to positive and negative weights respectively, and the total weight is 1.

[0024] Identify customer inquiry business keywords in the second semantic information and customer service answer business keywords in the first semantic information. Obtain the preset answer materials for the corresponding business based on the customer inquiry business. Calculate the matching degree between the customer service answer business keywords and the preset answer materials for the corresponding business to obtain the customer service business score.

[0025] Obtain the first sentiment keyword from the first semantic information, obtain the first sentiment score based on the sentiment positivity corresponding to the first sentiment keyword, obtain the second sentiment score based on the sentiment positivity corresponding to the first tone information, and obtain the first sentiment score based on the first sentiment score and the second sentiment score.

[0026] Obtain the second sentiment keywords from the second semantic information, obtain the first score of the second sentiment based on the sentiment positivity corresponding to the second sentiment keywords, obtain the second score of the second sentiment based on the sentiment positivity corresponding to the second tone information, and obtain the second sentiment score based on the first score of the second sentiment and the second score of the second sentiment.

[0027] Furthermore, based on at least one of the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score, service issues are identified and strategic suggestions are provided to guide customer service representatives in optimizing call services, specifically including:

[0028] If any of the customer service representative's communication skills score, business skills score, primary emotional score, or secondary emotional score falls below their preset score threshold, the service problem will be identified based on the score below the preset threshold, and corresponding strategy suggestions will be obtained. These strategy suggestions will be displayed to the customer service representative in real time during the call between the customer service representative and the customer to guide them in optimizing the call service.

[0029] Alternatively, based on the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score, a comprehensive score for the customer service call service can be obtained. If the comprehensive score is less than a preset comprehensive score threshold, the lowest score among the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score can be obtained. Based on the lowest score, the service problem can be located and corresponding strategy suggestions can be obtained. The corresponding strategy suggestions can be displayed to the customer service representative in real time during the call between the customer service representative and the customer to guide the customer service representative to optimize the call service.

[0030] Furthermore, based on the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score of the customer, a comprehensive score for the customer service call service is obtained, specifically including:

[0031] The overall score of customer service call service is obtained according to the following formula: P=aH+bY+cQ, where a, b, and c are the weights of the script rules, business rules, and emotional rules, respectively; H is the script score of the customer service representative; Y is the business score of the customer service representative; Q=(Q1+Q2) / 2 is the emotional score; Q1 is the first emotional score of the customer service representative; and Q2 is the second emotional score of the customer.

[0032] Furthermore, the system identifies service issues and obtains corresponding strategy suggestions. These suggestions are then displayed to customer service representatives in real-time during their calls to guide them in optimizing call service. Specifically, this includes:

[0033] If service issues are identified based on customer service script scores, keywords and prohibited phrases in customer service script rules can be obtained and highlighted to customer service representatives in real time, or phrases containing keywords in customer service script rules can be generated and provided to customer service representatives for reference in real time, so as to optimize customer service representatives' call services to customers.

[0034] If service issues are identified based on customer service performance scores, and customer service responses are generated based on pre-set information for customer inquiries and provided to customer service representatives in real time for reference, then customer service representatives can optimize their call service to customers.

[0035] If service issues are identified based on the customer service representative's initial emotional score, the representative can be prompted to adjust their own emotions in real time, and / or follow-up service statements composed of highly emotionally positive words can be generated and provided to the customer service representative for reference in real time, so as to optimize the customer service representative's call service to the customer;

[0036] If service issues are identified based on the customer's second emotional score, appropriate response statements can be generated to guide the customer in adjusting their emotions and provided to customer service representatives in real time to optimize customer service calls.

[0037] Secondly, this application provides a call center automation service optimization device, the device comprising:

[0038] The extraction module is used to acquire the voice recordings of customer service calls with customers, and extract the first semantic information and first tone information of the customer service representative and the second semantic information and second tone information of the customer from the voice recordings.

[0039] The scoring module, connected to the extraction module, is used to obtain the customer service's speech score, business score, first emotional score, and second emotional score based on the customer service's first semantic information, first tone information, and the customer's second semantic information, second tone information;

[0040] The guidance module, connected to the scoring module, is used to identify service issues and provide strategy suggestions to guide customer service representatives to optimize call services based on at least one of the customer service representative's script score, business score, primary emotional score, and secondary emotional score.

[0041] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the call center automation service optimization method as described above.

[0042] This application provides a method, apparatus, and medium for optimizing automated services in call centers. By distinguishing the semantics and tone in customer service and customer calls, it evaluates customer service and customer emotions from multiple dimensions to locate problems in call service and provides targeted guidance and strategy suggestions to optimize customer service call quality and improve customer satisfaction. Attached Figure Description

[0043] Figure 1 This is a flowchart of a call center automation service optimization method according to an embodiment of this application;

[0044] Figure 2 This is a flowchart of another call center automation service optimization method according to an embodiment of this application;

[0045] Figure 3 This is a schematic diagram of the structure of a call center automated service optimization device according to an embodiment of this application;

[0046] Figure 4 This is a schematic diagram of the structure of a computer-readable storage medium according to an embodiment of this application;

[0047] Figure 5 This is a schematic diagram of the structure of a computer device according to an embodiment of this application. Detailed Implementation

[0048] To enable those skilled in the art to better understand the technical solution of this application, the embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0049] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining this application and are not intended to limit this application.

[0050] It is understood that, without conflict, the various embodiments and features in the embodiments of this application can be combined with each other.

[0051] It is understood that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, while parts unrelated to this application are not shown in the drawings.

[0052] It is understood that each module or unit involved in the embodiments of this application may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple modules or units may be integrated into one entity structure.

[0053] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this application may occur in a different order than that marked in the accompanying drawings.

[0054] It is understood that the flowcharts and block diagrams of this application illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, unit, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagrams and flowcharts may be implemented using a hardware-based device to implement the specified function, or using a combination of hardware and computer instructions.

[0055] It is understood that the modules and units involved in the embodiments of this application can be implemented by software or by hardware. For example, the modules and units can be located in the processor.

[0056] Example 1:

[0057] like Figure 1 As shown, this application provides a method for optimizing call center automation services, the method comprising:

[0058] S1. Obtain the voice recording of the call between the customer service representative and the customer, and extract the first semantic information and first tone information of the customer service representative and the second semantic information and second tone information of the customer from the voice recording;

[0059] S2. Based on the customer service representative’s first semantic information, first tone information and the customer’s second semantic information, second tone information, obtain the customer service representative’s speech score, business score and first emotional score and the customer’s second emotional score.

[0060] S3. Based on at least one of the customer service representative's script score, business score, primary emotional score, and secondary emotional score, identify service issues and provide strategic suggestions to guide customer service representatives in optimizing call services.

[0061] In this embodiment, the provided method evaluates customer service and customer emotions from multiple dimensions by distinguishing between the semantics and tone in customer service and customer calls, in order to locate the problems in the call service and provide targeted guidance and strategy suggestions to optimize the quality of customer service calls and improve customer satisfaction.

[0062] Specifically, such as Figure 2As shown in the figure, this embodiment provides a method for optimizing call center automation services based on a large model.

[0063] The existing technology being addressed primarily involves customer service quality inspection methods that use natural language processing (NLP) to convert call voice data into text. Machine learning is then used to automatically analyze and evaluate the quality of customer service call recordings. However, these methods have several drawbacks: they use conventional, general-purpose speech recognition models to convert speech to text without specific optimization for particular application scenarios, resulting in lower speech recognition accuracy; while they inspect and provide feedback on call and service quality, they fail to offer guidance on how customer service representatives can improve service quality, hindering timely improvements; and the lack of noise reduction before speech recognition, or the use of only standard noise reduction methods, results in relatively low noise levels, impacting the accuracy of speech recognition.

[0064] In one embodiment, S1, acquiring the voice recording of a conversation between a customer service representative and a customer, and extracting the customer service representative's first semantic information and first tone information, and the customer's second semantic information and second tone information from the voice recording, specifically including:

[0065] During a call between customer service representatives and a customer, the original audio of the call is captured in real time.

[0066] The original call speech was denoised by combining Mel-frequency cepstral coefficients (MFCC) with a subtraction spectrum algorithm to obtain denoised call speech.

[0067] The noise-reduced call voice is input into the trained semantic tone recognition model to extract the customer service's first semantic information, first tone information, and second semantic information and second tone information from the call voice. The trained semantic tone recognition model consists of a trained language model and algorithms for recognizing speech rate and intonation.

[0068] In this embodiment, the large-model-based call center automation service optimization method is used in a terminal device. The large-model-based call center automation service optimization method includes the following steps:

[0069] S10 acquires voice data in real time, preprocesses the voice data to obtain processed voice data; specifically including:

[0070] S101, analyzes the connection status based on the call center, and determines whether the call is connected based on the connection status;

[0071] S102, if the call is not connected, only the unconnected number is recorded; if the call is connected, the voice signal after the connection is received and the voice data is obtained.

[0072] S103, extract speech data features, filter noise features based on speech data features, remove noise features, and obtain processed speech data.

[0073] By analyzing whether a call is connected, the nodes of speech recognition are adjusted to improve the recognition accuracy of speech data, extract speech data features, remove noise features, and reduce call interference.

[0074] In one embodiment, a combination of Mel-frequency cepstral coefficients (MFCC) and a subtraction spectral algorithm is used to denoise the original call speech to obtain denoised call speech, specifically including:

[0075] The Voice Activity Detection (VAD) algorithm is used to determine the speech segments and non-speech segments of the original call audio.

[0076] The power spectrum of the noisy speech segment and the noise power spectrum of the non-speech segment are obtained by using Mel frequency cepstral coefficients (MFCC).

[0077] The subtraction spectrum algorithm is used to subtract the noise power spectrum from the power spectrum of the noisy speech to obtain the power spectrum of the denoised speech and restore it to the denoised call speech.

[0078] In this embodiment, in step S103, a denoising method combining MFCC and subtraction spectrum algorithm is used. The specific denoising method is as follows:

[0079] Noise segment acquisition: The speech segment and non-speech segment (such as silence or pure noise segment) of the speech signal are determined by the speech activity detection (VAD) algorithm. The Mel Frequency Ceptral Coefficient (MFCC) feature of the noise is estimated in the non-speech segment.

[0080] Estimating the Mel frequency cepstral coefficients (MFCC) characteristics of noise specifically includes the following steps: The noise segment is divided into frames (e.g., 20ms / frame), and after pre-emphasis, Hanning windowing, FFT (Fast Fourier Transform), Mel filtering, and DCT (Discrete Cosine Transform), the noise MFCC of each frame is obtained.

[0081]

[0082] Calculate the mean and variance of noise MFCC: Perform statistical analysis on the MFCC features of non-speech segments, and calculate the mean and variance of each dimension:

[0083]

[0084]

[0085] The mean and variance of the noise MFCC are used for subsequent noise removal, where T is the number of frames in the non-speech segment and j is the dimension of the MFCC coefficient, which is usually taken as 12-13.

[0086] Noise power spectrum estimation:

[0087] Using IDCT (Inverse Discrete Cosine Transform) to... Reducing to the mean of the logarithmic Mel spectrum of the noise, and then exponentializing to obtain the linear Mel spectrum:

[0088]

[0089] Will Normalization yields the fluctuation weights for each Mel frequency band:

[0090]

[0091] Combined Mel filter response The noise power spectrum estimate is obtained as follows:

[0092]

[0093] Determine the subtraction spectral coefficients: Based on the MFCC characteristics of the speech signal and the noise signal, determine the subtraction spectral coefficients, which include the over-subtraction factor and the lower limit coefficient;

[0094] Over-subtraction factor:

[0095] Lower limit coefficient:

[0096] Frequency domain subtraction: Subtracting the spectrum of the noisy speech in the frequency domain yields the power spectrum of the denoised speech.

[0097]

[0098] in, The power spectrum of noisy speech is calculated using the following formula:

[0099]

[0100] Then, the power spectrum of the denoised speech is combined with the phase of the original noisy speech to obtain the denoised speech signal. The calculation formula is as follows:

[0101]

[0102] Overlap-preservation speech synthesis: The denoised speech signal is synthesized into a continuous speech signal by using the overlap-preservation method, thus restoring the time-domain waveform of the speech.

[0103] By extracting statistical features of noise (such as mean and variance) through MFCC, the noise spectrum can be located more accurately. Then, the subtraction spectrum algorithm is used to specifically remove noise, which is more efficient than a single algorithm. On the other hand, it is more suitable for call center scenarios: call center calls may contain environmental noise (such as background noise). This algorithm combines speech feature analysis and spectrum subtraction, which is suitable for the noise reduction needs of real-time calls, and provides cleaner speech data for subsequent semantic analysis (such as language large model processing).

[0104] In one implementation, the trained semantic tone recognition large model is obtained through the following steps:

[0105] A large language model is used as the base model. Algorithms for recognizing speech rate and intonation are added to the front end of the base model to obtain an initial large semantic tone recognition model.

[0106] A training dataset is formed by acquiring historical denoised call audio and corresponding historical semantic tone information. The historical semantic tone information includes historical first semantic information, historical first tone information, historical second semantic information and historical second tone information. The tone information is a number of levels that evaluate the call audio as good to bad.

[0107] The training dataset is input into the initial semantic tone recognition model for iterative training. This allows the initial semantic tone recognition model to first identify the speech rate and intonation of historical denoised call speech, and then obtain the recognition results based on the historical denoised call speech and its speech rate and intonation, and compare them with historical semantic tone information, until the training results converge to obtain the trained semantic tone recognition model.

[0108] In this embodiment, the call center automation service optimization method based on a large model includes the following steps:

[0109] S20, the processed speech data is input into the large language processing model (also known as the large language model LLM), the semantic information is output based on the large language processing model, and the semantic tone information is optimized by the constructed semantic tone optimization model, which is trained from historical speech data of the call center.

[0110] In step S20, the semantic tone information is optimized and adjusted by combining historical speech data, specifically including:

[0111] S201, obtain the matching degree between historical speech data and semantic tone information, and establish data pairs between historical speech data with a matching degree greater than or equal to a set matching degree threshold and semantic tone information;

[0112] S202, Based on the data, the initial semantic tone optimization model is iteratively trained on the established training set to obtain the training results;

[0113] S203, determine whether the training results have converged; convergence refers to the process of gradually approaching the optimal solution during training, and the loss function will gradually decrease until it reaches a low value, indicating that the semantic tone optimization model fits the training data better and better.

[0114] If convergence is achieved, a semantic mood optimization model is generated. Based on the semantic mood optimization model, the processed speech data is analyzed to obtain semantic mood information.

[0115] If convergence fails, adjust the matching threshold, generate a new training set, and dynamically adjust the model parameters of the semantic tone optimization model based on the new training set.

[0116] S204. Based on the semantic mood optimization model, the semantic mood information output by the large language processing model is optimized to obtain the optimized semantic mood information.

[0117] After the large language processing model outputs semantic information, a semantic mood optimization model is used to further optimize this information. This semantic mood optimization model is trained on historical speech data, thus it has a better understanding of the idiomatic expressions in this specific domain, and the optimized semantic mood information obtained through this model is more accurate.

[0118] The large-scale language processing model can be selected as needed, such as the GPT-4 large-scale language model, the DeepSeek large-scale language model, the Baidu Wenxin large-scale model, the Alibaba Tongyi large-scale model, the iFlytek Xinghuo large-scale model, etc. In practical applications, any large-scale language processing model can be used, or multiple large-scale language processing models can be used simultaneously. The outputs of multiple large-scale language processing models are fused using preset fusion rules to obtain relatively accurate semantic and tone information.

[0119] Tone information is used to assist in identifying emotions, and can be obtained by analyzing speech rate, intonation, and other factors. Both semantic and tone information distinguish between customer service representatives and customers, facilitating subsequent scoring across different dimensions.

[0120] In one embodiment, S2, based on the customer service representative's first semantic information and first tone information, and the customer's second semantic information and second tone information, obtains the customer service representative's script score, business score, first emotional score, and the customer's second emotional score, specifically including:

[0121] Obtain preset customer service script rule keywords and prohibited words. Based on the customer service script rule keywords and prohibited words contained in the first semantic information, and combined with the corresponding preset script word weights, calculate the customer service script score. The customer service script rule keywords and prohibited words are set to positive and negative weights respectively, and the total weight is 1.

[0122] Identify customer inquiry business keywords in the second semantic information and customer service answer business keywords in the first semantic information. Obtain the preset answer materials for the corresponding business based on the customer inquiry business. Calculate the matching degree between the customer service answer business keywords and the preset answer materials for the corresponding business to obtain the customer service business score.

[0123] Obtain the first sentiment keyword from the first semantic information, obtain the first sentiment score based on the sentiment positivity corresponding to the first sentiment keyword, obtain the second sentiment score based on the sentiment positivity corresponding to the first tone information, and obtain the first sentiment score based on the first sentiment score and the second sentiment score.

[0124] Obtain the second sentiment keywords from the second semantic information, obtain the first score of the second sentiment based on the sentiment positivity corresponding to the second sentiment keywords, obtain the second score of the second sentiment based on the sentiment positivity corresponding to the second tone information, and obtain the second sentiment score based on the first score of the second sentiment and the second score of the second sentiment.

[0125] In this embodiment, the call center automation service optimization method based on a large model includes the following steps:

[0126] S30 analyzes customer service's communication skills score, business skills score, emotional score, and customer emotional score based on optimized semantic and tone information.

[0127] The system has preset customer service-related script rules, business rules, and emotional rules. The script rules are the standard scripts used in the customer service process, the business rules are the process specifications for the business in the customer service process, and the emotional rules are the emotional requirements of the customer in the customer service process.

[0128] Based on optimized semantic information analysis, customer service script scores, business performance scores, and customer sentiment scores are calculated, specifically including:

[0129] S301, Semantic information is segmented using automatic sentence segmentation technology;

[0130] S302: Extract key words from customer service scripts, such as opening remarks, polite phrases, guiding phrases, prohibited phrases, and closing remarks. Based on the matching degree between the key words and the script rules, obtain the script score.

[0131] For example, the weights of various keywords can be preset, with a total weight of 1.0. Prohibited words have a negative weight, and the speech score is the weighted sum of the scores of each keyword category, as shown in Table 1 below:

[0132]

[0133] S303: Extract customer service business keywords, such as policy explanations, public service explanations, etc., and obtain a business score based on the matching degree between business keywords and business rules;

[0134] S304. Analyze the emotional information carried in the semantic information, including positive emotions such as happiness, praise, and gratitude, as well as negative emotions such as anger, resentment, and blame, to obtain the business customer sentiment score; score according to sentiment rules to obtain the customer sentiment score.

[0135] In one implementation, S3, based on at least one of the customer service representative's script score, business score, first emotional score, and second emotional score, identifies service issues and provides strategy suggestions to guide customer service representatives in optimizing call services, specifically including:

[0136] If any of the customer service representative's communication skills score, business skills score, primary emotional score, or secondary emotional score falls below their preset score threshold, the service problem will be identified based on the score below the preset threshold, and corresponding strategy suggestions will be obtained. These strategy suggestions will be displayed to the customer service representative in real time during the call between the customer service representative and the customer to guide them in optimizing the call service.

[0137] Alternatively, based on the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score, a comprehensive score for the customer service call service can be obtained. If the comprehensive score is less than a preset comprehensive score threshold, the lowest score among the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score can be obtained. Based on the lowest score, the service problem can be located and corresponding strategy suggestions can be obtained. The corresponding strategy suggestions can be displayed to the customer service representative in real time during the call between the customer service representative and the customer to guide the customer service representative to optimize the call service.

[0138] In this embodiment, the call center automation service optimization method based on a large model includes the following steps:

[0139] S40 transmits speech score, business score and customer sentiment score to the evaluation platform and obtains evaluation data in real time;

[0140] Based on the communication skills score, business skills score, and customer sentiment score, the data is transmitted to the evaluation platform, and evaluation data is obtained in real time, specifically including:

[0141] S401: Input the script score, business score, and customer sentiment score into the evaluation platform to obtain the evaluation value, or use the script score, business score, and customer sentiment score separately, set thresholds for each, and make corrections when the score is outside the threshold;

[0142] S402, determine whether the evaluation value is between the minimum evaluation threshold and the maximum evaluation threshold; the evaluation value must follow a normal distribution, that is, the overall evaluation value is between the preset minimum evaluation threshold and the maximum evaluation threshold. If it exceeds the threshold, it means that the call data may be abnormal and needs to be corrected. At this time, correction information is generated, and the abnormal state in the call data is compensated based on the correction information.

[0143] S403: If the evaluation value is between the minimum evaluation threshold and the maximum evaluation threshold, then obtain the evaluation data.

[0144] It should be noted that by evaluating customer service representatives and analyzing the evaluation values, it is beneficial to dynamically compensate for abnormal situations during the call and improve call quality.

[0145] S50: Based on the evaluation data analysis, customer satisfaction is assessed. When the satisfaction level is less than the satisfaction threshold, structured improvement information is generated and transmitted to the customer service terminal in real time. The structured improvement information is information that guides customer service to improve response information or response methods.

[0146] This method acquires and evaluates the conversations between customer service representatives and customers in real time to obtain customer satisfaction data. During the call, pop-up windows prompt customer service personnel so that they can adjust the conversation in a timely manner to improve customer satisfaction.

[0147] In one implementation, a comprehensive score for customer service call service is obtained based on the customer service representative's script score, business performance score, first emotional score, and the customer's second emotional score. Specifically, this includes:

[0148] The overall score of customer service call service is obtained according to the following formula: P=aH+bY+cQ, where a, b, and c are the weights of the script rules, business rules, and emotional rules, respectively; H is the script score of the customer service representative; Y is the business score of the customer service representative; Q=(Q1+Q2) / 2 is the emotional score; Q1 is the first emotional score of the customer service representative; and Q2 is the second emotional score of the customer.

[0149] In this embodiment, the specific evaluation value of S401 is: P=aH+bY+cQ, where a, b, and c are the weights of the script rules, business rules, and emotional rules, respectively. These weights can be preset and adjusted as needed. H, Y, and Q are the scores of the script rules, business rules, and emotional rules, respectively.

[0150] In one implementation, the service problem is located and corresponding strategy suggestions are obtained. These suggestions are then displayed to the customer service representative in real time during the call to guide them in optimizing the call service. Specifically, this includes:

[0151] If service issues are identified based on customer service script scores, keywords and prohibited phrases in customer service script rules can be obtained and highlighted to customer service representatives in real time, or phrases containing keywords in customer service script rules can be generated and provided to customer service representatives for reference in real time, so as to optimize customer service representatives' call services to customers.

[0152] If service issues are identified based on customer service performance scores, and customer service responses are generated based on pre-set information for customer inquiries and provided to customer service representatives in real time for reference, then customer service representatives can optimize their call service to customers.

[0153] If service issues are identified based on the customer service representative's initial emotional score, the representative can be prompted to adjust their own emotions in real time, and / or follow-up service statements composed of highly emotionally positive words can be generated and provided to the customer service representative for reference in real time, so as to optimize the customer service representative's call service to the customer;

[0154] If service issues are identified based on the customer's second emotional score, appropriate response statements can be generated to guide the customer in adjusting their emotions and provided to customer service representatives in real time to optimize customer service calls.

[0155] In this embodiment, step S50, analyzing customer satisfaction based on evaluation data and optimizing the service process in real time according to customer satisfaction, specifically includes:

[0156] S501, Obtain evaluation data, analyze customer feedback information based on the evaluation data, and analyze customer satisfaction based on the customer feedback data;

[0157] S502, determine whether customer satisfaction is greater than or equal to the set satisfaction threshold;

[0158] If the response is greater than or equal to the customer's, the call center's response is deemed to meet the customer's requirements.

[0159] If the response is less than the customer satisfaction level, the feedback information will be analyzed based on customer satisfaction. Based on the feedback information, the system will generate structured improvement information by matching the rule engine with the case library. This structured improvement information will be transmitted to the customer service client in real time, prompting them to improve the issues encountered during the customer service call based on the improvement information.

[0160] In step S502, the structured improvement information adopts a three-layer structured design of "problem location - strategy suggestion - operation example", specifically including:

[0161] Problem identification layer: Based on multi-dimensional scoring results, it can pinpoint the specific scenarios of service defects (such as "medical insurance reimbursement policy consultation", the type of violation (such as "incorrect interpretation of policy terms"), and the degree of impact (such as "leading to a 70% probability of citizens making incorrect applications").

[0162] Strategy recommendation layer: Provide 3-5 targeted and actionable improvement strategies, each strategy may include:

[0163] Script template: For example, when citizens inquire about the process of reimbursement for medical insurance in other places, the answer should be, "Hello, to reimburse medical insurance in other places, you need to first register with the medical insurance bureau in your place of insurance. After successful registration, you can directly settle the bill at the designated hospital in the place of treatment by taking your medical insurance card."

[0164] Business process guidance: For example, "Citizens need to be guided to log in to the 'National Medical Insurance Service Platform' APP (Application) for online registration. The operation path is: Open the APP → Click 'Recordation of Medical Treatment in Other Places' → Select the place of insurance and the place of medical treatment → Submit the registration application."

[0165] Emotional regulation techniques: For example, when a citizen speaks rapidly and is anxious, you can respond, "Don't worry, speak slowly, I will help you clarify the problem."

[0166] Operation Example Layer: Provides guidance on historical excellent call cases, such as providing links to recordings of excellent calls (e.g., "Call No. 20250308-1030, 5 minutes 15 seconds to 5 minutes 30 seconds") or the corresponding text dialogue records, as well as providing animated operation guidance for the government service system.

[0167] It should be noted that the structured improvement information implementation mechanism includes:

[0168] The differential feature extraction module identifies 28 types of service defect features from the evaluation data (such as "incorrect interpretation of policies", "omission of procedures", and "cold service attitude").

[0169] Strategy generation engine: A generative model built on 80,000+ historical excellent service cases, supporting accurate mapping of "government service defects - improvement strategies";

[0170] Interactive Presentation System: Utilizing a dedicated pop-up window format for government services, an interactive improvement guidance layer is overlaid on the customer service interface in real-time. The guidance content is linked to the service standard terminology database. The interactive presentation format is as follows: an orange warning box pops up in the upper right corner of the screen, labeled "XX Business Consultation Service Optimization Suggestion," containing bold text guidance, a standard terminology pronunciation playback button, and a highlighted dynamic guidance animation showing the operation path of the government service system. Clicking "Detailed Solution" expands the improvement plan, including links to relevant policy documents, comparative analysis of historical excellent service cases, and real-time annotations of the dialogue content with citizens, clearly indicating areas for improvement, such as highlighting ambiguous expressions in red and providing annotations of standardized terminology.

[0171] It should be noted that by analyzing customer satisfaction, the call center's response information is dynamically adjusted in real time to ensure that the call center's responses meet customer requirements and improve service quality.

[0172] This large-model-based method for optimizing call center automation services includes the following steps:

[0173] Step S60: After the call ends, store the call data and corresponding semantic tone information for use as a new training set to further train the semantic tone optimization model, so as to improve the semantic tone optimization model.

[0174] It should be noted that the semantic tone optimization model can be further trained whenever new data is added to the training set. Since too many new training sets are added daily, the system overhead from frequent training is significant. To address this issue, periodic training can be scheduled, such as daily or weekly training using newly added training sets to optimize the semantic tone model. This reduces the system overhead caused by excessive training and saves system resources.

[0175] The beneficial effects of this embodiment include:

[0176] 1) After the semantic information is output by the large language processing model, the semantic tone optimization model is used to optimize the semantic tone information. This semantic tone optimization model is trained based on historical voice data (such as historical voice data of government hotlines). Therefore, this semantic tone optimization model is more familiar with the common language in this specific field. The optimized semantic tone information obtained by this semantic tone optimization model is also more accurate, and more accurate semantic information can be obtained, thereby better understanding customer needs.

[0177] 2) Structured improvement information generation and interaction: Based on customer satisfaction analysis and feedback discrepancies, structured improvement information is generated through a rule engine and case library matching. This structured improvement information is transmitted to the customer service end in real time, prompting them to address issues during customer service calls. This structured improvement information includes a three-level content architecture: problem identification, strategy suggestions, and operation examples. It provides solutions from quality feedback to precise problem-solving, enabling real-time quality checks and the transmission of improvement information to the customer service end during calls. This helps customer service representatives improve service quality in a targeted manner and enhance customer satisfaction.

[0178] 3) After the call ends, store the call data and corresponding semantic information for use as a new training set to further train the semantic optimization model, thereby improving the semantic optimization model. Set up regular training, such as daily or weekly, to use the new training set to train the semantic optimization model, so as to reduce the system overhead caused by too many training times and save system resources.

[0179] In addition, in terms of economic benefits, it can reduce operating costs, improve customer service efficiency, increase revenue opportunities, and enhance customer loyalty; in terms of social benefits, it can improve user experience, enhance the industry image, and promote the digitalization of society.

[0180] Example 2:

[0181] like Figure 3 As shown, this application provides a call center automation service optimization device, the device comprising:

[0182] Extraction module 1 is used to acquire the voice recordings of customer service and customer calls, and extract the first semantic information and first tone information of the customer service representative and the second semantic information and second tone information of the customer from the voice recordings.

[0183] The scoring module 2, connected to the extraction module 1, is used to obtain the customer service's speech score, business score, first emotional score, and second emotional score based on the customer service's first semantic information, first tone information, and the customer's second semantic information, second tone information;

[0184] The guidance module 3, connected to the scoring module 2, is used to locate service problems and provide strategy suggestions to guide customer service to optimize call services based on at least one of the customer service's script score, business score, first emotion score, and the customer's second emotion score.

[0185] In one embodiment, the extraction module 1 specifically includes:

[0186] The raw unit is used to acquire the raw audio of the call between the customer service representative and the customer in real time.

[0187] The denoising unit, connected to the original unit, is used to denoise the original call speech by combining Mel frequency cepstral coefficients (MFCC) with a subtraction spectrum algorithm to obtain denoised call speech.

[0188] The extraction unit, connected to the denoising unit, is used to input the denoised call speech into the trained semantic tone recognition model to extract the customer service's first semantic information, first tone information, and second semantic information and second tone information from the call speech. The trained semantic tone recognition model consists of a trained language model and algorithms for recognizing speech rate and intonation.

[0189] In one embodiment, the noise reduction unit specifically includes:

[0190] The VAD unit is used to determine the speech segments and non-speech segments of the original call speech through the speech activity detection VAD algorithm;

[0191] The MFCC unit, connected to the VAD unit, is used to obtain the noisy speech power spectrum of the speech segment and the noise power spectrum of the non-speech segment using Mel-frequency cepstral coefficients (MFCC).

[0192] The subtraction spectrum unit, connected to the MFCC unit, is used to subtract the noise power spectrum from the noisy speech power spectrum using the subtraction spectrum algorithm to obtain the denoised speech power spectrum and recover the denoised call speech.

[0193] In one embodiment, the trained semantic tone recognition model in the extraction unit is obtained through the following steps:

[0194] A large language model is used as the base model. Algorithms for recognizing speech rate and intonation are added to the front end of the base model to obtain an initial large semantic tone recognition model.

[0195] A training dataset is formed by acquiring historical denoised call audio and corresponding historical semantic tone information. The historical semantic tone information includes historical first semantic information, historical first tone information, historical second semantic information and historical second tone information. The tone information is a number of levels that evaluate the call audio as good to bad.

[0196] The training dataset is input into the initial semantic tone recognition model for iterative training. This allows the initial semantic tone recognition model to first identify the speech rate and intonation of historical denoised call speech, and then obtain the recognition results based on the historical denoised call speech and its speech rate and intonation, and compare them with historical semantic tone information, until the training results converge to obtain the trained semantic tone recognition model.

[0197] In one embodiment, the scoring module 2 specifically includes:

[0198] The script scoring unit is used to obtain preset customer service script rule keywords and prohibited words. Based on the customer service script rule keywords and prohibited words contained in the first semantic information, combined with the corresponding preset script word weights, the customer service script score is calculated. The customer service script rule keywords and prohibited words are set to positive and negative weights respectively, and the total weight is 1.

[0199] The business scoring unit is used to identify customer inquiry business in the second semantic information and customer service answer business keywords in the first semantic information. Based on the customer inquiry business, it obtains the preset answer materials for the corresponding business and calculates the matching degree between the customer service answer business keywords and the preset answer materials for the corresponding business to obtain the customer service business score.

[0200] The sentiment scoring unit is used to obtain the first sentiment keyword from the first semantic information, obtain a first sentiment score based on the sentiment positivity corresponding to the first sentiment keyword, obtain a second sentiment score based on the sentiment positivity corresponding to the first tone information, and obtain a first sentiment score based on the first sentiment score and the second sentiment score; and,

[0201] Obtain the second sentiment keywords from the second semantic information, obtain the first score of the second sentiment based on the sentiment positivity corresponding to the second sentiment keywords, obtain the second score of the second sentiment based on the sentiment positivity corresponding to the second tone information, and obtain the second sentiment score based on the first score of the second sentiment and the second score of the second sentiment.

[0202] In one embodiment, the guidance module 3 specifically includes:

[0203] The first guidance unit is used to locate service problems and obtain corresponding strategy suggestions based on the score of the customer service representative, the business score, the first emotion score, and the second emotion score of the customer if any of these scores is lower than its own preset score threshold. The corresponding strategy suggestions are then displayed to the customer service representative in real time during the call between the customer service representative and the customer to guide the customer service representative to optimize the call service.

[0204] Alternatively, the second guidance unit is used to obtain a comprehensive score for customer service call service based on the customer service representative's script score, business score, first emotional score, and the customer's second emotional score. If the comprehensive score is less than a preset comprehensive score threshold, the unit obtains the lowest score among the customer service representative's script score, business score, first emotional score, and the customer's second emotional score, locates the service problem based on the lowest score, obtains the corresponding strategy suggestion, and displays the corresponding strategy suggestion to the customer service representative in real time during the call between the customer service representative and the customer to guide the customer service representative to optimize the call service.

[0205] In one embodiment, the second guidance unit specifically includes a comprehensive scoring unit, specifically used for:

[0206] The overall score of customer service call service is obtained according to the following formula: P=aH+bY+cQ, where a, b, and c are the weights of the script rules, business rules, and emotional rules, respectively; H is the script score of the customer service representative; Y is the business score of the customer service representative; Q=(Q1+Q2) / 2 is the emotional score; Q1 is the first emotional score of the customer service representative; and Q2 is the second emotional score of the customer.

[0207] In one embodiment, the guidance module 3 specifically includes:

[0208] The first positioning suggestion optimization unit is used to locate service problems based on customer service script scores, obtain customer service script rule keywords and prohibited words and provide them to customer service in real time, or generate script phrases containing customer service script rule keywords and provide them to customer service for reference in real time, so as to optimize customer service's call service to customers.

[0209] The second positioning suggestion optimization unit is used to locate service problems based on the customer service's business score, generate customer service response statements based on the preset answer materials of customer consultation business and provide them to customer service for reference in real time, so as to optimize the customer service's call service to customers.

[0210] The third positioning suggestion optimization unit is used to locate service problems based on the customer service's first emotional score, prompt the customer service to adjust their own emotions in real time, and / or generate follow-up service statements composed of words with high emotional positivity and provide them to the customer service for reference in real time, so as to optimize the customer service's call service to customers.

[0211] The fourth positioning suggestion optimization unit is used to locate service issues based on the customer's second emotion score, generate response statements to guide the customer to adjust their emotions, and provide them to customer service for reference in real time, so as to optimize the customer service's call service to the customer.

[0212] Example 3:

[0213] like Figure 4 As shown, Embodiment 3 of this application provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it implements the call center automation service optimization method as described in Embodiment 1.

[0214] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program units, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0215] like Figure 5 As shown, this application can also provide a computer device, including a memory and a processor. The memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the call center automation service optimization method as described in Embodiment 1. This computer device can be the call center automation service optimization device as described in Embodiment 2.

[0216] The memory is connected to the processor. The memory can be flash memory, read-only memory or other types of memory. The processor can be a central processing unit or a microcontroller.

[0217] Embodiments 1-3 of this application provide a method, apparatus, and medium for optimizing automated services in call centers. By distinguishing the semantics and tone in customer service and customer calls, the method evaluates customer service and customer emotions from multiple dimensions to locate problems in call services and provides targeted guidance and strategy suggestions to optimize the quality of customer service calls and improve customer satisfaction.

[0218] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of this application, and this application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this application, and these modifications and improvements are also considered to be within the scope of protection of this application.

Claims

1. A method for optimizing automated services in a call center, characterized in that, The method includes: Obtain the voice recordings of conversations between customer service representatives and customers, and extract the primary semantic information and primary tone information of the customer service representatives, as well as the secondary semantic information and secondary tone information of the customers from the voice recordings. Based on the customer service representative's first semantic information and first tone information, and the customer's second semantic information and second tone information, obtain the customer service representative's speech score, business score, first emotional score, and the customer's second emotional score; Based on at least one of the customer service representative's communication skills score, business skills score, primary emotional score, and secondary emotional score, identify service issues and provide strategic suggestions to guide customer service representatives in optimizing call services.

2. The method according to claim 1, characterized in that, Obtain the voice recordings of conversations between customer service representatives and customers, and extract the primary semantic information and primary tone information of the customer service representative, as well as the secondary semantic information and secondary tone information of the customer. Specifically, this includes: During a call between customer service representatives and a customer, the original audio of the call is captured in real time. The original call speech was denoised by combining Mel-frequency cepstral coefficients (MFCC) with a subtraction spectrum algorithm to obtain denoised call speech. The noise-reduced call voice is input into the trained semantic tone recognition model to extract the customer service's first semantic information, first tone information, and second semantic information and second tone information from the call voice. The trained semantic tone recognition model consists of a trained language model and algorithms for recognizing speech rate and intonation.

3. The method according to claim 2, characterized in that, The method combines Mel-frequency cepstral coefficients (MFCC) with a subtraction spectral algorithm to denoise the original call speech, specifically including: The Voice Activity Detection (VAD) algorithm is used to determine the speech segments and non-speech segments of the original call audio. The power spectrum of the noisy speech segment and the noise power spectrum of the non-speech segment are obtained by using Mel frequency cepstral coefficients (MFCC). The subtraction spectrum algorithm is used to subtract the noise power spectrum from the power spectrum of the noisy speech to obtain the power spectrum of the denoised speech and restore it to the denoised call speech.

4. The method according to claim 2, characterized in that, The large-scale semantic tone recognition model after training is obtained through the following steps: A large language model is used as the base model. Algorithms for recognizing speech rate and intonation are added to the front end of the base model to obtain an initial large semantic tone recognition model. A training dataset is formed by acquiring historical denoised call audio and corresponding historical semantic tone information. The historical semantic tone information includes historical first semantic information, historical first tone information, historical second semantic information and historical second tone information. The tone information is a number of levels that evaluate the call audio as good to bad. The training dataset is input into the initial semantic tone recognition model for iterative training. This allows the initial semantic tone recognition model to first identify the speech rate and intonation of historical denoised call speech, and then obtain the recognition results based on the historical denoised call speech and its speech rate and intonation, and compare them with historical semantic tone information, until the training results converge to obtain the trained semantic tone recognition model.

5. The method according to claim 1, characterized in that, Based on the customer service representative's primary semantic information and primary tone information, and the customer's secondary semantic information and secondary tone information, obtain the customer service representative's communication skills score, business skills score, primary emotional score, and the customer's secondary emotional score, specifically including: Obtain preset customer service script rule keywords and prohibited words. Based on the customer service script rule keywords and prohibited words contained in the first semantic information, and combined with the corresponding preset script word weights, calculate the customer service script score. The customer service script rule keywords and prohibited words are set to positive and negative weights respectively, and the total weight is 1. Identify customer inquiry business keywords in the second semantic information and customer service answer business keywords in the first semantic information. Obtain the preset answer materials for the corresponding business based on the customer inquiry business. Calculate the matching degree between the customer service answer business keywords and the preset answer materials for the corresponding business to obtain the customer service business score. Obtain the first sentiment keyword from the first semantic information, obtain the first sentiment score based on the sentiment positivity corresponding to the first sentiment keyword, obtain the second sentiment score based on the sentiment positivity corresponding to the first tone information, and obtain the first sentiment score based on the first sentiment score and the second sentiment score. Obtain the second sentiment keywords from the second semantic information, obtain the first score of the second sentiment based on the sentiment positivity corresponding to the second sentiment keywords, obtain the second score of the second sentiment based on the sentiment positivity corresponding to the second tone information, and obtain the second sentiment score based on the first score of the second sentiment and the second score of the second sentiment.

6. The method according to any one of claims 1-5, characterized in that, Based on at least one of the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score, identify service issues and provide strategic suggestions to guide customer service representatives in optimizing call services, specifically including: If any of the customer service representative's communication skills score, business skills score, primary emotional score, or secondary emotional score falls below their preset score threshold, the service problem will be identified based on the score below the preset threshold, and corresponding strategy suggestions will be obtained. These strategy suggestions will be displayed to the customer service representative in real time during the call between the customer service representative and the customer to guide them in optimizing the call service. Alternatively, based on the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score, a comprehensive score for the customer service call service can be obtained. If the comprehensive score is less than a preset comprehensive score threshold, the lowest score among the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score can be obtained. Based on the lowest score, the service problem can be located and corresponding strategy suggestions can be obtained. The corresponding strategy suggestions can be displayed to the customer service representative in real time during the call between the customer service representative and the customer to guide the customer service representative to optimize the call service.

7. The method according to claim 6, characterized in that, A comprehensive score for customer service call service is obtained based on the customer service representative's script score, business performance score, primary emotional score, and secondary emotional score of the customer. This score includes: The overall score of customer service call service is obtained according to the following formula: P=aH+bY+cQ, where a, b, and c are the weights of the script rules, business rules, and emotional rules, respectively; H is the script score of the customer service representative; Y is the business score of the customer service representative; Q=(Q1+Q2) / 2 is the emotional score; Q1 is the first emotional score of the customer service representative; and Q2 is the second emotional score of the customer.

8. The method according to claim 6, characterized in that, Locate service issues and obtain corresponding strategy suggestions. These suggestions are then displayed to customer service representatives in real-time during their calls to guide them in optimizing call service. Specifically, this includes: If service issues are identified based on customer service script scores, keywords and prohibited phrases in customer service script rules can be obtained and highlighted to customer service representatives in real time, or phrases containing keywords in customer service script rules can be generated and provided to customer service representatives for reference in real time, so as to optimize customer service representatives' call services to customers. If service issues are identified based on customer service performance scores, and customer service responses are generated based on pre-set information for customer inquiries and provided to customer service representatives in real time for reference, then customer service representatives can optimize their call service to customers. If service issues are identified based on the customer service representative's initial emotional score, the representative can be prompted to adjust their own emotions in real time, and / or follow-up service statements composed of highly emotionally positive words can be generated and provided to the customer service representative for reference in real time, so as to optimize the customer service representative's call service to the customer; If service issues are identified based on the customer's second emotional score, appropriate response statements can be generated to guide the customer in adjusting their emotions and provided to customer service representatives in real time to optimize customer service calls.

9. A call center automated service optimization device, characterized in that, The device includes: The extraction module is used to acquire the voice recordings of customer service calls with customers, and extract the first semantic information and first tone information of the customer service representative and the second semantic information and second tone information of the customer from the voice recordings. The scoring module, connected to the extraction module, is used to obtain the customer service's speech score, business score, first emotional score, and second emotional score based on the customer service's first semantic information, first tone information, and the customer's second semantic information, second tone information; The guidance module, connected to the scoring module, is used to identify service issues and provide strategy suggestions to guide customer service representatives to optimize call services based on at least one of the customer service representative's script score, business score, primary emotional score, and secondary emotional score.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the call center automation service optimization method as described in any one of claims 1-8.

Images