Customer service training method, device, equipment and medium

Abstract

The invention discloses a customer service training method and device, equipment and a medium. The method comprises the steps of determining first prompt word information of a first round of partner training dialogue according to information to be trained, and inputting the first prompt word information into a pre-trained verbal skill generation model to obtain customer consultation simulation information of the first round of partner training dialogue; response information of the customer service staff to the customer consultation simulation information is determined; adjusting the cue word information according to the response information and / or the customer consultation simulation information to obtain updated cue word information of the next round of partner training dialogue; and inputting the update prompt word information into the verbal skill generation model to obtain updated customer consultation simulation information of the next round of partner training conversation, and determining update response information of the customer service staff to the updated customer consultation simulation information until the training of the customer service staff is finished. According to the invention, the prompt word information is dynamically adjusted according to different response information of customer service, the training verbal skill is automatically adjusted according to different customer service training conditions, and the partner training effect and accuracy of training are improved.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a customer service training method, device, equipment and medium. Background Art

[0002] With the development of e-commerce, customer service (hereinafter referred to as "customer service") has been applied to various fields, such as technical support, after-sales service, and banking applications. Customer service is a key channel of communication between customers and businesses, and satisfactory customer service can greatly enhance corporate value. Therefore, excellent customer service personnel who can solve customer problems are crucial. To improve the business capabilities and technical level of customer service personnel, companies conduct customer service training on a regular or irregular basis. Because manual training makes it difficult to track each trainee's learning progress and training results in real time, and training costs are high, with the development of artificial intelligence, the use of intelligent robots for customer service training is gradually becoming mainstream.

[0003] An intelligent customer service training system uses an intelligent conversational robot to train or coach customer service personnel. The robot plays the role of the customer, and the trainee plays the customer service representative. During the training process, the robot simulates real customer requests and seeks help from the training staff, who are then tasked with resolving the customer's issue as best they can. Traditional intelligent customer service training systems require instructors to design training script templates, script flow, and customer intent. This requires instructors to have strong business experience and script flow design expertise. Furthermore, the scripts used by intelligent training robots based on script flow design are relatively simple and cannot automatically adjust and adapt to the individual training student's practice, resulting in limited effectiveness. Summary of the Invention

[0004] The present invention provides a customer service training method, device, equipment and medium to solve the problem of low efficiency of customer service training.

[0005] According to one aspect of the present invention, a customer service training method is provided, comprising:

[0006] Determining first prompt word information for a first round of training dialogue based on the information to be trained, and inputting the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue;

[0007] Determine the customer service's response information to the customer consultation simulation information;

[0008] Adjust the prompt word information according to the response information or the response information and the customer consultation simulation information to obtain updated prompt word information for the next round of training dialogue;

[0009] The updated prompt word information is input into the speech generation model to obtain the updated customer consultation simulation information of the next round of training dialogue, and the updated response information of the customer service to the updated customer consultation simulation information is determined. The updated prompt word information of the next round of training dialogue is repeatedly determined based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

[0010] According to another aspect of the present invention, there is provided a customer service training device, comprising:

[0011] A first-round consultation speech generation module is used to determine first prompt word information for the first round of training dialogue based on the information to be trained, and input the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue;

[0012] A first-round response information determination module is used to determine the customer service's response information to the customer consultation simulation information;

[0013] a prompt word dynamic adjustment module, configured to adjust the prompt word information according to the response information or the response information and the customer consultation simulation information, to obtain updated prompt word information for the next round of training dialogue;

[0014] The dynamic speech generation module is used to input the updated prompt word information into the speech generation model, obtain the updated customer consultation simulation information of the next round of training dialogue, and determine the updated response information of the customer service to the updated customer consultation simulation information, and repeatedly determine the updated prompt word information of the next round of training dialogue based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

[0015] According to another aspect of the present invention, an electronic device is provided, comprising:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the customer service training method described in any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the customer service training method described in any embodiment of the present invention when executed.

[0020] The technical solution of the embodiment of the present invention solves the problem of single speech through a speech generation model, and dynamically adjusts the prompt word information of the speech generation model according to the different response information of the customer service, so as to realize automatic adjustment and adaptation of the training speech according to different customer service training situations, provide the training customer service with more accurate and more humane training speech, and improve the training effect and accuracy.

[0021] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 is a flow chart of a customer service training method provided according to an embodiment of the present invention;

[0024] Figure 2 is a flow chart of another customer service training method provided according to an embodiment of the present invention;

[0025] Figure 3 is a flow chart of another customer service training method provided according to an embodiment of the present invention;

[0026] Figure 4 This is a structural diagram of a customer service training system provided according to an embodiment of the present invention;

[0027] Figure 5 This is a structural diagram of a prompt word framework unit provided according to an embodiment of the present invention;

[0028] Figure 6 This is a structural diagram of a customer service training device provided according to an embodiment of the present invention;

[0029] Figure 7 The figure is a schematic diagram of the structure of an electronic device for implementing the customer service training method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "candidate", "target", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0032] Figure 1 A flowchart of a customer service training method is provided for an embodiment of the present invention. This embodiment is applicable to situations where a machine using a configuration model is used to train customer service. The method can be executed by a customer service training device, which can be implemented in the form of hardware and / or software. The customer service training device can be configured in a server with computing capabilities. Figure 1 As shown, the method includes:

[0033] S110 , determining first prompt word information for the first round of training dialogue based on the information to be trained, and inputting the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue.

[0034] The information to be trained includes relevant information that needs to be provided for customer service training. For example, the information to be trained includes training topic information. The training topic information is the customer problem that needs to be solved in this training, for example, solving the problem of customer bank card password being locked.

[0035] The speech generation model is a Large Language Model (LLM), a deep learning model trained using large amounts of text data. It can generate natural language text or understand the meaning of text. Large language models can handle a variety of natural language tasks, such as text classification, question-answering, and conversation, and are an important path to artificial intelligence. Currently, large language models use a similar Transformer architecture and pre-training objectives (such as Language Modeling) as small models. The difference between large language models and small models lies in the increased model size, training data, and computing resources.

[0036] Prompt information is the large model's prompt, used for interacting with the speech generation model. It can be text or voice, and its primary function is to guide the speech generation model in responding to specific inputs. The design and selection of prompts is crucial to ensuring that the speech generation model's output aligns with real customer inquiries. Prompts serve as the trigger text for the large language model's response, guiding the speech generation model's response. Their purpose is to guide the speech generation model in a specific way to meet user needs or complete specific tasks. Prompts can improve interaction efficiency, enhance the relevance and quality of generated content, and promote learning and creative thinking.

[0037] The speech generation model acts as a trainer for customer service representatives. The model outputs simulated customer inquiries and assesses the customer service performance during these simulated customer-to-customer interactions. The model simulates real customers, raising questions to the customer service representatives, acting as a sparring partner and assisting them in their practice.

[0038] Specifically, the initial prompt word information in the first round of training dialogue is determined based on the training topic information in the information to be trained, and is used as the first prompt word information. The first prompt word information is used to guide the speech generation model to determine the customer's consultation information in different scenarios. For example, the first prompt word information is a simulated customer's first consultation speech to solve the problem of the customer's bank card password being locked. The first prompt word information is input into the speech generation model, and the model outputs customer consultation simulation information that meets the requirements based on the first prompt word information. For example, the customer consultation simulation information is "Hello, I forgot my bank card password and it is now locked. What should I do?"

[0039] In a feasible embodiment, determining the first prompt word information of the first round of training dialogue according to the information to be trained includes:

[0040] The role prompt words and the theme prompt words in the prompt word template are filled in according to the role simulation information and the training theme information in the information to be trained, so as to obtain the first prompt word information.

[0041] A prompt word template is predetermined, and the prompt word template includes at least role prompt words and topic prompt words. For example, the prompt word template is "As an XX role, in order to solve XX problem, output consulting words", where XX role is the role prompt word and solving XX problem is the topic prompt word. According to the role simulation information and training topic information in the information to be trained, the corresponding prompt words in the prompt word template are filled in, and the first prompt word information is obtained after filling in. The information to be trained is determined according to the training needs, and semantic recognition is performed on the information to be trained to obtain role simulation information and training topic information. The role simulation information and training topic information are added to the prompt word template to obtain the first prompt word information.

[0042] S120: Determine the customer service's response information to the customer inquiry simulation information.

[0043] The simulated customer inquiry information output by the speech generation model is displayed to the customer service representative. The customer service representative treats the simulated customer inquiry information as real customer input, answers the questions in the simulated customer inquiry information, and obtains the customer service representative's response information. Specifically, a speech synthesis model is used to convert the simulated customer inquiry information into audio information and play it back to the customer service representative. The customer service representative responds as a voice message. The speech recognition model is used to recognize the customer service representative's response audio and convert it into text as the response information.

[0044] S130: Adjust the prompt word information according to the response information or the response information and the customer consultation simulation information to obtain updated prompt word information for the next round of training dialogue.

[0045] The current customer service training progress is determined based on the response information and / or customer consultation simulation information, and the prompt words are adjusted based on the current customer service training progress to obtain updated prompt word information for the next round of training dialogue. The current customer service training progress is used to represent the customer service level information, for example, the current customer service training progress includes service professional parameters and service emotions, etc.

[0046] The customer service representative's sentiment is determined based on the response, and the prompts are adjusted accordingly to generate updated prompts for the next round of training. For example, the updated prompts might be, "The customer service representative's patience is currently 80%. As a consulting customer, I need to continue providing consulting advice to resolve the issue of my bank card password being locked."

[0047] The customer service representative's response accuracy is determined by combining the response information and the simulated customer consultation information. This accuracy represents how accurately the customer service representative answered the questions in the simulated customer consultation information. This accuracy is used as a service professional parameter, and the prompt word information is adjusted based on the service professional parameter to obtain updated prompt word information for the next round of training dialogue. For example, the updated prompt word information may be "The current customer service representative's answer accuracy for this training topic question is 80%. As a consulting customer, in order to resolve the bank card password lock issue, continue to output the consultation words."

[0048] In one feasible embodiment, S130 includes:

[0049] Determine historical context information based on response information and customer inquiry simulation information;

[0050] The historical conversation background prompt words in the update prompt word information are determined according to the historical context background information.

[0051] Obtain the historical response information and historical customer consultation simulation information in the current round of practice dialogue, perform semantic recognition on all historical response information and historical customer consultation simulation information, use the semantic recognition results as historical context background information, and update the historical dialogue context prompt words in the prompt word information according to the historical context background information to assist the speech generation model in determining the current dialogue progress and topic resolution progress.

[0052] This embodiment determines the contextual background information through the currently updated training dialogue, and dynamically updates the prompt word information based on the contextual background information, so that the updated prompt word information includes the aforementioned dialogue background information, thereby improving the accuracy of the next round of customer consultation simulation information output by the speech generation model.

[0053] S140. Input the updated prompt word information into the speech generation model to obtain the updated customer consultation simulation information for the next round of training dialogue, and determine the customer service's updated response information to the updated customer consultation simulation information. Repeat the determination of the updated prompt word information for the next round of training dialogue based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

[0054] This updated prompt is fed back into the speech generation model, which then generates updated customer inquiry simulation information based on the updated prompt. This updated customer inquiry simulation information is then displayed to the customer service representative, who then uses it as a real customer's question, answers the questions in the updated customer inquiry simulation information, and continues to receive updated responses from the customer service representative.

[0055] The updated prompt word information for the next training round is dynamically determined based on the customer service representative's updated responses and simulated customer inquiries. This ensures that the customer inquiry simulation information output by the speech generation model during each training round is dynamically adjusted based on the customer service representative's status, improving the accuracy of the customer inquiry simulation information output. If the preset training round number is reached, the prompt word information is stopped and the training is concluded. Alternatively, if the customer service representative has fully answered all questions on the training topic based on the responses from previous training sessions and simulated customer inquiries, the training is concluded.

[0056] The technical solution of this embodiment solves the problem of single speech through a speech generation model, and dynamically adjusts the prompt word information of the speech generation model according to the different response information of the customer service, so as to realize automatic adjustment and adaptation of the training speech according to different customer service training situations, provide more accurate and more humane training speech for the training customer service, and improve the training effect and accuracy.

[0057] Figure 2 This is a flow chart of another customer service training method provided by an embodiment of the present invention. This embodiment further refines the dynamic adjustment of the prompt words in the above embodiment. Figure 2 As shown, the method includes:

[0058] S210: Determine first prompt word information for the first round of training dialogue based on the information to be trained, and input the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue.

[0059] S220: Determine the customer service's response information to the customer inquiry simulation information.

[0060] S230: Perform emotion recognition based on the response information to obtain target customer service emotion information.

[0061] A text sentiment classification model is used to perform sentiment recognition on the response information to obtain the target customer service sentiment information. The text sentiment classification model of this embodiment uses the RoBerta model to perform text sentiment recognition and obtain the sentiment classification result as the target customer service sentiment information. Specifically comprising: Step 1: Convert the response information into text information, input it into the RoBerta model Embedding module after preprocessing, encode it into a word embedding vector, a sentence segment embedding vector, and a position encoding vector, and add the three vectors to form the Embedding output; Step 2: Send the vector obtained in Step 1 to the Encoder module of the Transformer to encode the feature vector; Step 3: After the feature vector obtained in Step 2 is processed by the fully connected layer, it is normalized using the Softmax function to obtain the sentiment classification result of the current text as the target customer service sentiment information. The sentiment classification result can be a sentiment score, for example, 1 to 10 points. The lower the score, the worse the sentiment. The details of the specific score can be adjusted according to the actual situation and are not limited here.

[0062] The RoBerta model is a Transformer-based pre-trained language model with the same structure as the BERT model. RoBerta improves model performance through larger datasets, better data preprocessing, and improved learning rate adjustment. It can be applied to a variety of NLP tasks, such as text classification, named entity recognition, and sentiment analysis. Embedding refers to the process of converting a certain type of input data (such as text, images, or sounds) into a dense numerical vector. This vector typically contains multiple dimensions, each of which represents an abstract feature or attribute of the input data.

[0063] Furthermore, the response information includes response voice information, and a speech emotion classification model is used to perform emotion recognition on the response voice information to obtain a first emotion classification result; a text emotion classification model is used to perform emotion recognition on the response text information to obtain a second emotion classification result; and target customer service emotion information is obtained based on the first emotion classification result and the second emotion classification result. Exemplarily, the speech emotion classification model is a speech recognition model, and the specific type of speech recognition model is not limited. The first emotion classification result is a first emotion score, and the second emotion classification result is a second emotion score. The target customer service emotion information is determined based on the weighted value of the first emotion score and the second emotion score. The first weight corresponding to the first emotion score and the second weight corresponding to the second emotion score can be determined based on the actual scenario requirements and are not limited here.

[0064] Since voice also includes some human emotions, combining text information and voice information to determine customer service emotions can improve the accuracy of customer service emotion recognition.

[0065] S240: Determine target customer emotion information corresponding to the target customer service emotion information based on a predetermined mapping relationship between customer service emotions and customer emotions.

[0066] The mapping relationship between customer service agent emotions and customer emotions includes a correspondence between multiple candidate customer service agent emotions and multiple candidate customer emotions, for example, a correspondence between multiple candidate customer service agent emotion scores and multiple candidate customer emotion scores. Within this mapping relationship, target customer emotion information that matches the target customer service agent emotion information is determined to simulate how real customers might react to customer service agent emotions in subsequent conversations, thereby enhancing the authenticity of the training exercise and, consequently, the effectiveness of customer service training.

[0067] In a feasible embodiment, the process of determining the mapping relationship between customer service emotions and customer emotions includes:

[0068] Obtain historical consultation dialogue information in real consultation scenarios; wherein the historical consultation dialogue information includes multiple historical customer consultation information and corresponding historical customer service response information;

[0069] Perform emotion recognition on each historical customer consultation information and historical customer service response information respectively to obtain multiple historical customer service emotion information and corresponding historical customer emotion information;

[0070] Modeling is performed based on multiple historical customer service sentiment information and corresponding historical customer sentiment information to obtain a mapping relationship between customer service sentiment and customer sentiment.

[0071] Among them, the historical consultation dialogue information in the real consultation scenario refers to the consultation dialogue information between real customers and customer service.

[0072] Emotion recognition is performed on each historical customer consultation information and historical customer service response information to obtain corresponding historical customer service emotion information and historical customer emotion information. The identification of emotion information can refer to the above steps and will not be repeated here.

[0073] Since the next round of historical customer consultation information is fed back based on the previous round of historical customer service response information, the historical customer sentiment information of the next round of historical customer consultation information corresponds to the historical customer service sentiment information of the previous round of historical customer service response information.

[0074] Modeling is performed based on historical customer service emotion information and historical customer emotion information that have a corresponding relationship, and the mapping relationship between customer service emotion and customer emotion is determined based on the corresponding relationship. For example, the specific modeling method can be determined based on fuzzy matching, that is, the fuzzy matching method is used to match multiple historical customer service emotion information and the corresponding historical customer emotion information to obtain the mapping relationship between customer service emotion and customer emotion.

[0075] This embodiment improves the accuracy of determining the mapping relationship by identifying historical customer emotion information and historical customer service emotion information in real consultation scenarios and determining the corresponding relationship, and then obtaining a mapping relationship based on the historical customer emotion information and historical customer service emotion information with the corresponding relationship.

[0076] S250: Determine the emotional prompt words in the updated prompt word information according to the target customer's emotional information.

[0077] The target customer's emotional information is used as the emotional prompt word in the updated prompt word information, so that when the speech generation model determines the updated customer consultation simulation information based on the updated prompt word information, the updated customer consultation simulation information can more accurately simulate the emotional changes brought about by the real customer based on the aforementioned response of the customer service.

[0078] S260. Input the updated prompt word information into the speech generation model to obtain the updated customer consultation simulation information for the next round of training dialogue, and determine the customer service's updated response information to the updated customer consultation simulation information. Repeat the determination of the updated prompt word information for the next round of training dialogue based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

[0079] The technical solution of this embodiment determines the customer service emotion through the customer service response information, and dynamically adjusts the customer emotion in the prompt word information according to the changes in the customer service emotion, so as to timely adjust the training direction according to the response of different customer service staff and improve the training effect.

[0080] Figure 3 This is a flow chart of another customer service training method provided by an embodiment of the present invention. This embodiment further refines the dynamic adjustment of the prompt words in the above embodiment. Figure 3 As shown, the method includes:

[0081] S310: Determine first prompt word information for the first round of training dialogue based on the information to be trained, and input the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue.

[0082] S320: Determine the customer service's response information to the customer inquiry simulation information.

[0083] S330. Determine at least one target standard consultation information based on the matching results between the customer consultation simulation information and a plurality of candidate standard consultation information in the standard script library.

[0084] Among them, the standard script library includes multiple candidate standard consultation information and corresponding candidate standard response information under predetermined different training topics. The standard script library can be determined based on industry experience and is not limited here.

[0085] A similarity calculation is performed between the simulated customer inquiry information and multiple candidate standard inquiry information corresponding to the same training topic. The similarity between the simulated customer inquiry information and each candidate standard inquiry information is obtained as a matching result. The candidate standard inquiry information with a similarity greater than a similarity threshold is selected as the target standard inquiry information. The similarity threshold can be determined based on actual scenarios, and the specific value of the threshold is not limited herein.

[0086] Exemplarily, multiple candidate standard consultation information in the standard speech library is input into the BGE M3-Embedding text embedding model to be encoded into text vectors, thereby obtaining multiple candidate standard consultation vectors; the customer consultation simulation information is also input into the BGE M3-Embedding text embedding model to be encoded into text vectors, thereby obtaining a simulated consultation vector; the cosine similarity between the simulated consultation vector and each candidate standard consultation vector is calculated as the matching result. The candidate standard consultation information with the largest cosine similarity is selected as the target standard consultation information. Among them, the BGE M3-Embedding model is used to create an advanced embedding type of "learning-based sparse embedding". The advantage of this embedding is that it combines the accuracy of sparse embedding and the semantic richness of dense embedding.

[0087] S340. Determine the gap parameter between the response information and the standard speech based on the matching result of the target standard response information and the response information corresponding to at least one target standard consultation information in the standard speech library.

[0088] Determine the candidate standard response information corresponding to the target standard consultation information in the standard speech library as the target standard response information, determine the similarity between the response information and the target standard response information, and use the similarity as the gap parameter between the response information and the standard speech. Exemplarily, the target standard response information and the response information are input into the BGE M3-Embedding text embedding model to encode them into text vectors, obtaining a target standard response vector and a response vector, calculating the cosine similarity between each target standard response vector and the response vector as a matching result, and using the value of the cosine similarity as the value of the gap parameter between the response information and the standard speech.

[0089] Since the target standard consultation information is similar to the customer consultation simulation information, according to the standard customer service level, the response information of the customer service in response to the customer consultation simulation information should be similar to the target standard response information corresponding to the target standard consultation information. The gap parameter between the response information and the target standard response information reflects the correctness of the customer service's current response.

[0090] S350: Determine, according to the gap parameter, the training direction prompt word in the updated prompt word information.

[0091] The training direction prompt is used to determine the content direction of the next training dialogue, and also to determine the number of rounds of the next training dialogue.

[0092] Specifically, the correspondence between different gap parameters and different training directions is determined in advance, such as the correspondence between different gap parameters and the remaining training dialogue rounds, and the matching remaining training dialogue rounds are determined according to the gap parameters, thereby guiding the speech generation model to determine the next consultation speech according to the remaining training dialogue rounds, ensuring the integrity of the dialogue between all training dialogue rounds, and avoiding the situation where the corresponding customer consultation simulation information at the end of the customer service training does not conform to the communication logic of the end of the consultation, and is still in the process of communicating on the training topic, affecting the training effect.

[0093] In one feasible embodiment, S350 includes:

[0094] Determine the remaining training dialogue rounds and the degree of deviation from the training topic based on the gap parameters;

[0095] The training direction prompt words in the updated prompt word information are determined based on the remaining training dialogue rounds and the degree of deviation from the training topic.

[0096] Predetermine in advance the round mapping relationship between different candidate gap parameters and candidate remaining sparring dialogue rounds, as well as the deviation degree mapping relationship between different candidate gap parameters and candidate training topic deviation degrees, and determine the matching remaining sparring dialogue rounds in the round mapping relationship based on the current gap parameters, and determine the matching training topic deviation degree in the deviation degree mapping relationship.

[0097] The remaining training conversation rounds and the degree of deviation from the training topic are used as training direction prompts in the updated prompt information. For example, the training direction prompt in the updated prompt information is "The remaining training conversation rounds are 5 rounds, the degree of deviation from the training topic is moderate, in order to solve the problem of bank card password being locked, continue to output consultation scripts." This is to guide the script generation model to make the next customer consultation simulation information more in line with the training topic according to the remaining training conversation rounds and the degree of deviation from the training topic in the training direction prompt, so as to avoid being misled by the incorrect response information of the customer service in the previous round, causing the subsequent customer consultation simulation information to deviate further from the training topic.

[0098] This embodiment updates the prompt word information based on the remaining training dialogue rounds and the degree of deviation from the training topic, further ensuring the accuracy of the customer consultation simulation information determined based on the prompt word information.

[0099] S360. Input the updated prompt word information into the speech generation model to obtain the updated customer consultation simulation information for the next round of training dialogue, and determine the customer service's updated response information to the updated customer consultation simulation information. Repeat the determination of the updated prompt word information for the next round of training dialogue based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

[0100] The technical solution of this embodiment determines the training direction prompt words in the prompt word information by determining the gap between the customer service's response information and the standard speech, so as to timely adjust the training direction according to the response situation of different customer service staff and improve the training effect; at the same time, it can accurately evaluate the accuracy of the customer service's reply and determine the customer service training results.

[0101] Figure 4 This is a structural diagram of a customer service training system provided by an embodiment of the present invention. Figure 4 As shown, the system includes: a prompt word framework unit, a training speech generation unit, a speech synthesis unit, a student response unit and a speech recognition unit.

[0102] The prompt word framework unit adjusts prompt word information based on the response information or the response information and customer inquiry simulation information, generating updated prompt word information for the next round of training conversations. This prompt word information can be dynamically adjusted based on the training status of each customer service representative, assisting the speech generation model in generating training scripts more tailored to the individual customer service representative. The training speech generation unit primarily uses the speech generation model to simulate real customers, pose business questions to the customer service representative, acting as a training instructor and assisting the customer service representative in practice. The speech synthesis unit converts the customer inquiry simulation information generated by the speech generation model into an audio stream. The student response unit plays the customer inquiry simulation information in audio mode while simultaneously capturing the customer service representative's response audio. The speech recognition unit converts the captured customer service representative response audio into text to generate the response information. The overall workflow of this system is shown below.

[0103] Step 1: Initialize the training topic, emotion setting, and goal completion status, and fill in the prompt word template to obtain the first prompt word information. Step 2: Input the filled first prompt word information into the speech generation model, which generates the training speech as the customer consultation simulation information. Step 3: Use the speech synthesis model to convert the training speech generated in Step 2 into audio. Step 4: After the customer service representative listens to the training audio, they make a voice response. The speech recognition model recognizes the student's audio and converts it into text to obtain the response information. Step 5: Based on the response information or the response information and the customer consultation simulation information, the prompt word information is adjusted to obtain the updated prompt word information for the next round of training dialogue. Step 6: Repeat steps 2-5 above until the training goal is achieved.

[0104] Furthermore, the prompt word framework unit is mainly used to adjust the prompt word information transmitted to the speech generation model. The unit is mainly composed of a static prompt word management module and a dynamic prompt word management module. The static prompt word management module includes three parts: a role setting device, a theme setting device, and a context filling device; the dynamic prompt word management module includes an emotion setting device and a training direction instruction setting device. Figure 5 Shown is a schematic structural diagram of the prompt word framework unit.

[0105] The role-setting component provides background information for the speech generation model. The model is set to play the role of the customer, while the customer service representative is the person answering the customer's questions. During the training process, customer issues are resolved in a question-and-answer format. The content set in the role-setting component is static and generally does not need to be modified during training.

[0106] The topic setting component specifies the training task for the speech generation model. This task specifies the customer problem to be solved, such as resolving a customer's bank card password lockout issue. The topic setting component's settings are populated when the training task is initiated and typically do not need to be modified during the training process.

[0107] The context-filling component provides the speech generation model with contextual information, primarily historical conversations between the model and the customer service representative, to assist the model in understanding the semantics of the speech generation model. This information is continuously updated during the training process as the conversation content changes.

[0108] The emotion setting device consists of an emotion recognition sub-device and an emotion mapping sub-device. The emotion recognition sub-device uses a text-based sentiment classification model to identify changes in the customer service representative's emotions during the training process. This information is then fed into the emotion mapping sub-device. This sub-device, fed with the customer service representative's emotions, outputs the set customer emotions, simulating the different emotional changes a real customer might experience based on the customer service representative's emotional reactions.

[0109] The training direction setting component is used to determine the discrepancy between customer service responses and standard scripts, thereby timely adjusting the training direction of the script generation model. This component consists of a knowledge base sub-component, a script retrieval sub-component, and a command mapping sub-component.

[0110] The knowledge base sub-component stores the standard script library corresponding to the training topic. This paper uses a vector database to store candidate standard inquiry information and candidate standard response information for this training topic. The script retrieval sub-component uses cosine similarity to calculate the similarity score between the conversation data between the script generation model and the customer and the standard scripts in the knowledge base sub-component. This score is input to the instruction mapping sub-component to obtain the training direction instruction for the current task. The specific steps executed by the training direction instruction setting component are as follows.

[0111] Step 1: Predetermine the candidate standard consultation information and candidate standard response information corresponding to the training topic; Step 2: Input the candidate standard consultation information and candidate standard response information into the BGE M3-Embedding text embedding model to encode it into a text vector, obtain the standard speech text vector, and store it in the knowledge base sub-device; Step 3: Obtain the current customer consultation simulation information and response information of this training task, and input it into the BGE The M3-Embedding text embedding model determines the training text vector; Step 4: Calculate the cosine similarity between the training text vector and the standard speech text vector stored in the knowledge base sub-device to determine the highest similarity score; Step 5: Compare the highest similarity score with the preset training task completion threshold. If the threshold condition is reached, the training task ends; Step 6: Compare the current interaction rounds between the training customer service and the speech generation model with the pre-set maximum interaction rounds. If the maximum interaction rounds have been reached, the training task ends; Step 7: If both steps 5 and 6 are not met, the highest similarity score is input as the gap parameter to the instruction mapping sub-device to determine the training direction instruction corresponding to the highest similarity score.

[0112] An embodiment of the present invention proposes a large model prompt word framework specifically for intelligent customer service training scenarios. This prompt word framework can guide the large model to generate richer, more effective, and more comprehensive training scripts; through the emotion setting device in this embodiment, the emotion setting of the large model can be updated in real time according to the emotional changes of the customer service, so that the large model can generate more humanistic and emotional training scripts; through the training direction instruction setting device in the embodiment of the present invention, the current training effect of the customer service can be evaluated in real time, and the trainee's practice accuracy and learning status can be judged in real time; and the large model training direction can be updated in real time according to the current training effect of the customer service, so that the training effect can be more controllable and efficient.

[0113] Figure 6 This is a structural diagram of a customer service training device provided by an embodiment of the present invention. Figure 6 As shown, the device includes:

[0114] A first-round consultation speech generation module 610 is configured to determine first prompt word information for the first round of training dialogue based on the information to be trained, and input the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue;

[0115] A first-round response information determination module 620 is used to determine the customer service's response information to the customer consultation simulation information;

[0116] The prompt word dynamic adjustment module 630 is used to adjust the prompt word information according to the response information or the response information and the customer consultation simulation information to obtain updated prompt word information for the next round of training dialogue;

[0117] The dynamic speech generation module 640 is used to input the updated prompt word information into the speech generation model, obtain the updated customer consultation simulation information of the next round of training dialogue, and determine the updated response information of the customer service to the updated customer consultation simulation information, and repeatedly determine the updated prompt word information of the next round of training dialogue based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

[0118] The technical solution of this embodiment solves the problem of single speech through a speech generation model, and dynamically adjusts the prompt word information of the speech generation model according to the different response information of the customer service, so as to realize automatic adjustment and adaptation of the training speech according to different customer service training situations, provide more accurate and more humane training speech for the training customer service, and improve the training effect and accuracy.

[0119] Optionally, the prompt word dynamic adjustment module includes a first dynamic adjustment unit, specifically configured to:

[0120] Perform emotion recognition based on the response information to obtain target customer service emotion information;

[0121] Determining target customer emotion information corresponding to the target customer emotion information based on a predetermined mapping relationship between customer service emotions and customer emotions;

[0122] The emotional prompt word in the updated prompt word information is determined according to the target customer emotional information.

[0123] Optionally, the device further includes a mapping relationship determination module, configured to determine a mapping relationship between the customer service emotion and the customer emotion, specifically configured to:

[0124] Obtaining historical consultation dialogue information in a real consultation scenario; wherein the historical consultation dialogue information includes multiple historical customer consultation information and corresponding historical customer service response information;

[0125] Performing emotion recognition on each of the historical customer consultation information and the historical customer service response information to obtain a plurality of historical customer service emotion information and corresponding historical customer emotion information;

[0126] Modeling is performed based on the multiple historical customer service emotion information and the corresponding historical customer emotion information to obtain a mapping relationship between customer service emotion and customer emotion.

[0127] Optionally, the prompt word dynamic adjustment module includes a second dynamic adjustment unit, including:

[0128] a consultation information matching subunit, configured to determine at least one target standard consultation information based on a matching result between the customer consultation simulation information and a plurality of candidate standard consultation information in a standard script library;

[0129] a gap parameter determination subunit, configured to determine a gap parameter between the response information and the standard speech based on a matching result between the target standard response information corresponding to the at least one target standard consultation information in the standard speech library and the response information;

[0130] The training direction determining subunit is configured to determine the training direction prompt word in the updated prompt word information according to the gap parameter.

[0131] Optionally, the training direction determination subunit is specifically used to:

[0132] determining the remaining training dialogue rounds and the degree of deviation from the training topic according to the gap parameters;

[0133] The training direction prompt word in the updated prompt word information is determined according to the remaining training dialogue rounds and the degree of deviation from the training topic.

[0134] Optionally, the first-round consultation script generation module includes a first prompt word information determination unit, which is specifically used to:

[0135] The role prompt words and the theme prompt words in the prompt word template are filled in according to the role simulation information and the training theme information in the information to be trained, so as to obtain the first prompt word information.

[0136] Optionally, the prompt word dynamic adjustment module includes a third dynamic adjustment unit, specifically configured to:

[0137] Determining historical context information based on the response information and the customer consultation simulation information;

[0138] The historical conversation context prompt words in the update prompt word information are determined according to the historical context background information.

[0139] The customer service training device provided in the embodiment of the present invention can execute the customer service training method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0140] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations and do not violate public order and good morals.

[0141] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0142] Figure 7A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0143] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0144] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0145] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the customer service training method.

[0146] In some embodiments, the customer service training method can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the customer service training method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the customer service training method in any other suitable manner (e.g., via firmware).

[0147] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific reference products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0148] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0149] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0150] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0151] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes switch components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, switch components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0152] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0153] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product that includes a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication unit 19, or installed from the storage unit 18, or installed from the ROM 12. When the computer program is executed by the processor 11, the above-mentioned functions defined in the method of the embodiment of the present invention are performed.

[0154] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0155] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A customer service training method, characterized in that: The method includes: Determining first prompt word information for a first round of training dialogue based on the information to be trained, and inputting the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue; Determine the customer service's response information to the customer consultation simulation information; Adjust the prompt word information according to the response information or the response information and the customer consultation simulation information to obtain updated prompt word information for the next round of training dialogue; The updated prompt word information is input into the speech generation model to obtain the updated customer consultation simulation information of the next round of training dialogue, and the updated response information of the customer service to the updated customer consultation simulation information is determined. The updated prompt word information of the next round of training dialogue is repeatedly determined based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

2. The method according to claim 1, characterized in that Adjusting the prompt word information according to the response information to obtain updated prompt word information for the next round of training dialogue includes: Perform emotion recognition based on the response information to obtain target customer service emotion information; Determining target customer emotion information corresponding to the target customer emotion information based on a predetermined mapping relationship between customer service emotions and customer emotions; The emotional prompt word in the updated prompt word information is determined according to the target customer emotional information.

3. The method according to claim 2, characterized in that in, The process of determining the mapping relationship between customer service emotions and customer emotions includes: Obtaining historical consultation dialogue information in a real consultation scenario; wherein the historical consultation dialogue information includes multiple historical customer consultation information and corresponding historical customer service response information; Performing emotion recognition on each of the historical customer consultation information and the historical customer service response information to obtain a plurality of historical customer service emotion information and corresponding historical customer emotion information; Modeling is performed based on the multiple historical customer service emotion information and the corresponding historical customer emotion information to obtain a mapping relationship between customer service emotion and customer emotion.

4. The method according to claim 1, wherein The prompt word information is adjusted according to the response information and the customer consultation simulation information to obtain updated prompt word information for the next round of training dialogue, including: Determining at least one target standard consultation information based on a matching result between the customer consultation simulation information and a plurality of candidate standard consultation information in a standard script library; determining a gap parameter between the response information and the standard speech based on a matching result between the target standard response information corresponding to the at least one target standard consultation information and the response information in the standard speech library; The training direction prompt word in the updated prompt word information is determined according to the gap parameter.

5. The method according to claim 4, characterized in that Determining the sparring direction prompt word in the updated prompt word information according to the gap parameter includes: determining the remaining training dialogue rounds and the degree of deviation from the training topic according to the gap parameters; The training direction prompt word in the updated prompt word information is determined according to the remaining training dialogue rounds and the degree of deviation from the training topic.

6. The method according to any one of claims 1 to 5, characterized in that The first prompt word information of the first round of training dialogue is determined based on the information to be trained, including: The role prompt words and the theme prompt words in the prompt word template are filled in according to the role simulation information and the training theme information in the information to be trained, so as to obtain the first prompt word information.

7. The method according to any one of claims 1 to 5, characterized in that The prompt word information is adjusted according to the response information and the customer consultation simulation information to obtain updated prompt word information for the next round of training dialogue, including: Determining historical context information based on the response information and the customer consultation simulation information; The historical conversation context prompt words in the update prompt word information are determined according to the historical context background information.

8. A customer service training device, characterized in that: The device includes: A first-round consultation speech generation module is used to determine first prompt word information for the first round of training dialogue based on the information to be trained, and input the first prompt word information into a pre-trained speech generation model to obtain customer consultation simulation information for the first round of training dialogue; A first-round response information determination module is used to determine the customer service's response information to the customer consultation simulation information; a prompt word dynamic adjustment module, configured to adjust the prompt word information according to the response information or the response information and the customer consultation simulation information, to obtain updated prompt word information for the next round of training dialogue; The dynamic speech generation module is used to input the updated prompt word information into the speech generation model, obtain the updated customer consultation simulation information of the next round of training dialogue, and determine the updated response information of the customer service to the updated customer consultation simulation information, and repeatedly determine the updated prompt word information of the next round of training dialogue based on the response information and customer consultation simulation information of the previous round of training dialogue until the training of the customer service is completed.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to perform the customer service training method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the customer service training method according to any one of claims 1 to 7 when executed.

Images