Method and device for determining verbal skill template and program product

Through multiple iterations and experimental customer service evaluation, the speech template is adjusted, and the problem of incomplete speech template of the machine customer service is solved, the business completion degree and user experience are improved, and faster iteration convergence and accurate evaluation are achieved.

CN119047447APending Publication Date: 2024-11-29WEBANK (CHINA)
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411066404.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The existing machine customer service has problems such as incomplete settings of speech templates and does not meet the needs of target customers, resulting in low business completion and poor user experience.

Method used

Adjust the speech template through the M iteration process, use experimental customer service and generative big model training to evaluate and expand the call content, determine the optimal speech template, achieve smaller granularity comparison and rapid convergence, and meet the needs of target customers.

Benefits of technology

Improve business completion and user experience, reduce subjective factors of manual evaluation, and achieve faster iterative convergence and more accurate evaluation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047447A_ABST
    Figure CN119047447A_ABST
Patent Text Reader

Abstract

The invention provides a verbal skill template determination method and device and a program product, and the method comprises the steps: adjusting a verbal skill template through M iteration processes, thereby determining an optimal verbal skill template, and the ith iteration process comprises the steps: distributing target customers with the same business attribute in the ith period to a plurality of experimental customer services in the ith period; determining a verbal skill template corresponding to the experimental customer service with the highest service completion degree in the ith period according to call contents of different experimental customer services in the ith period; judging whether any customer service verbal skill in the verbal skill template corresponding to the experimental customer service with the highest service completion degree in the ith period has a sub-verbal skill or not; if yes, expanding the verbal skill template corresponding to the experimental customer service with the highest business completion degree in the ith period according to the sub-verbal skill, and obtaining a first business model according to the sub-verbal skill; and according to the first service model and the plurality of service models of the ith period, determining that the plurality of experimental customer services of the (i + 1) th period enter the (i + 1) th iteration process. According to the scheme, the service completion degree and the user experience can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device and program product for determining a speech template. Background Art

[0002] With the development of artificial intelligence technology, more and more companies are using machine customer service instead of human customer service to provide services, thereby saving labor costs, increasing business processing volume and meeting user needs.

[0003] However, machine customer service is not as flexible as human customer service and is more dependent on speech templates; good speech templates can not only improve the user experience, but also improve the completion of business.

[0004] At present, most speech templates are set according to the designer's experience. Human experience is limited, and different designers have different ideas. Problems such as incomplete speech templates and speech templates not meeting the needs of target customers may occur, resulting in low business completion and poor user experience.

[0005] So, how to set up speech templates to improve business completion and user experience remains to be solved. Summary of the invention

[0006] The present application provides a method, device and program product for determining a speech template, which can improve the completion of business and user experience.

[0007] In the first aspect, the embodiment of the present application provides a method for determining a speech template, which can be executed by a speech template determination device, and the speech template determination device can be a terminal device or a module for a terminal device, or a server or a module for a server. The present application does not limit the execution subject of the method. The method includes: adjusting the speech template through M iterations to determine the optimal speech template, M is a positive integer greater than 2, and the i-th iteration process includes: assigning target customers with the same business attributes in the i-th period to multiple experimental customer services in the i-th period; the business model of any experimental customer service is obtained by inputting each speech template into each generative large model for training; wherein different experimental customer services have different business models; i is a positive integer; receiving the call content of different experimental customer services in the i-th period; the call content is obtained by the different experimental customer services in the i-th period talking with the corresponding target customers according to their own business models; according to the different experimental customer services in the i-th period Based on the call content, determine the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period; determine whether any customer service speech in the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period has a sub-speech; if so, expand the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the sub-speech; input the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the sub-speech into the generative large model for training to obtain a first business model; based on the first business model and multiple business models of the i-th period, determine that multiple experimental customer services in the i+1th period enter the i+1th iteration process.

[0008] In the above scheme, on the one hand, the business model of the experimental customer service is obtained by inputting each speech template into each generative large model for training. Compared with manual customer service, it is easier to expand the number of experimental customer service. Since each experimental customer service corresponds to at least one speech template, the speech template can be divided into smaller granularity, so that the business completion of different speech templates can be better compared, and faster iterative convergence can be achieved. However, the number and energy of manual customer service are limited. If the speech template is divided according to manual customer service, the granularity of the speech template is large. The business completion of the speech template may be good in a certain period, but the business completion of the speech template in the next period is poor, with large volatility and difficulty in convergence. On the other hand, the speech template corresponding to the experimental customer service with the highest business completion is divided into smaller granularity, which can achieve rapid convergence. On the other hand, the evaluation and expansion of the speech template in each period are obtained based on the actual communication content with the target customer, which can meet the needs of the target customer, thereby improving the business completion and user experience.

[0009] In one possible implementation method, it is determined whether there is a new intention in the call content of different experimental customer service personnel in the i-th period; if so, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded according to the new intention; the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the new intention are input into the generative large model for training to obtain a second business model; and multiple experimental customer services in the i+1th period are determined according to the first business model, the second business model and multiple business models of the i-th period.

[0010] The above scheme, on the one hand, can divide the speech template into smaller granularities according to the new intent, so as to achieve better comparison of the business completion of different speech templates and achieve faster iterative convergence; on the other hand, each new intent is determined according to the specific call content, that is, the new intent is provided by the target customer and can reflect the real needs of the target customer, so that the expanded business template fits the needs of the target customer, thereby improving business completion and user experience.

[0011] In one possible implementation method, the call contents of different experimental customer service staff in the i-th period are input into the quality inspection model to obtain the quality inspection result of each experimental customer service; based on the quality inspection result of each experimental customer service, the experimental customer service with the highest business completion rate in the i-th period is determined; based on the call contents of the experimental customer service with the highest business completion rate in the i-th period, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is determined.

[0012] In the above scheme, on the one hand, the quality inspection model can quickly and effectively evaluate the call content of different experimental customer service staff, especially when the call content is large, which is more efficient than manual evaluation; on the other hand, the quality inspection model can achieve a relatively accurate evaluation of the call content of different experimental customer service staff, reducing the influence of subjective factors in manual evaluation.

[0013] In one possible implementation method, the call content of each call made by the experimental customer service with the highest business completion rate in the i-th period is obtained; the call content of each call is scored to obtain the scoring result of each call; and the call content of the call with the highest scoring result is determined as the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period.

[0014] In the above scheme, since the content of each call made by the experimental customer service may be different from the training speech template corresponding to the experimental customer service, the training speech template corresponding to the experimental customer service can be adjusted according to the content of the call with the highest score, so that the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th cycle is more in line with the actual application scenario, and further, it can improve business completion and user experience.

[0015] In one possible implementation method, the speech content of each call is scored based on the hit rate of the speech, the degree of match between the speech and the preset template, and / or the context relevance; the tone of voice is scored based on the rationality of the tone and / or the harmlessness of the tone; the intention response score is scored based on the rationality of the intended response sequence, the degree of match between the intended response sequence and the preset template and / or the number of intended responses; and the scoring result for each call is obtained based on the speech score, the tone score and the intended response score.

[0016] The above scheme can accurately and effectively determine the scoring result of each call based on the speech scoring, tone scoring and intention response scoring.

[0017] In a possible implementation method, the generative big model is fine-tuned according to the quality inspection role, speech scoring rules, tone scoring rules and intention scoring rules to determine the quality inspection big model; the quality inspection role is used to indicate that the role type of the generative big model is quality inspection.

[0018] In the above solution, the generative big model is a generative model based on a large-scale corpus, which makes the natural language understanding and generation of the quality inspection big model closer to the human level and can improve the accuracy of quality inspection.

[0019] In a possible implementation method, the target customer of the i-th period is assigned to at least two experimental customer service staff and at least one manual customer service staff of the i-th period; during the conversation with the target customer, the manual customer service staff can adaptively adjust the speech template according to the business attributes.

[0020] In the above scheme, the manual customer service has strong adaptive ability, can respond according to the needs of different target customers, and adjust the speech template according to the call content of the manual customer service, so that the adjusted business template can fit the needs of the target customers, thereby improving business completion and user experience.

[0021] In one possible implementation method, the speech template is adjusted through M iterative processes to determine the optimal speech template, including: determining the speech template of the experimental customer service with the highest business completion rate in the Mth period as the optimal speech template; or, when the experimental customer service with the highest business completion rate is the same for N consecutive times, determining the speech template of the experimental customer service with the highest business completion rate as the optimal speech template; wherein N is a positive integer greater than 2.

[0022] The above scheme can accurately and effectively determine the optimal speech template.

[0023] In a second aspect, an embodiment of the present application provides a device for determining a speech template, including: a transceiver unit and a processing unit. The transceiver unit is used to assign target customers with the same business attributes in the i-th period to multiple experimental customer services in the i-th period; the business model of any experimental customer service is obtained by inputting each speech template into each generative large model for training; wherein different experimental customer services have different business models; i is a positive integer; receiving the call content of different experimental customer services in the i-th period; the call content is obtained by the different experimental customer services in the i-th period talking with the corresponding target customers according to their own business models; the processing unit is used to determine the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the call content of different experimental customer services in the i-th period; judging whether any customer service speech in the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period has a sub-speech; if so, expanding the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the sub-speech; inputting the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the sub-speech into the generative large model for training to obtain a first business model; determining that multiple experimental customer services in the i+1th period enter the i+1th iteration process according to the first business model and the multiple business models in the i-th period.

[0024] In a possible implementation method, the processing unit is used to determine whether there is a new intention in the call content of different experimental customer service personnel in the i-th period; if so, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded according to the new intention; the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the new intention are input into the generative large model for training to obtain a second business model; and multiple experimental customer services for the i+1th period are determined according to the first business model, the second business model and multiple business models of the i-th period.

[0025] In one possible implementation method, the processing unit is used to input the call contents of different experimental customer services in the i-th period into the quality inspection big model to obtain the quality inspection result of each experimental customer service; determine the experimental customer service with the highest business completion rate in the i-th period based on the quality inspection result of each experimental customer service; and determine the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period based on the call contents of the experimental customer service with the highest business completion rate in the i-th period.

[0026] In one possible implementation method, the processing unit is used to obtain the call content of each call made by the experimental customer service with the highest business completion rate in the i-th period; score the call content of each call to obtain the scoring result of each call; and determine the call content of the call with the highest scoring result as the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period.

[0027] In one possible implementation method, the processing unit is used to score the speech content of each phone call based on the hit rate of the speech, the degree of match between the speech and a preset template, and / or the context relevance; to score the tone of the call content of each phone call based on the rationality of the tone and / or the harmlessness of the tone; to score the intention response of the call content of each phone call based on the rationality of the intention response sequence, the degree of match between the intention response sequence and a preset template and / or the number of intention responses; and to obtain a scoring result for each phone call based on the speech score, the tone score and the intention response score.

[0028] In one possible implementation method, the processing unit is used to fine-tune the generative big model according to the quality inspection role, speech scoring rules, tone scoring rules and intention scoring rules to determine the quality inspection big model; the quality inspection role is used to indicate that the role type of the generative big model is quality inspection.

[0029] In a possible implementation method, the transceiver unit is used to assign the target customer of the i-th period to at least two experimental customer service staff and at least one manual customer service staff of the i-th period; during the conversation with the target customer, the manual customer service staff can adaptively adjust the speech template according to the business attributes.

[0030] In one possible implementation method, the processing unit is used to determine the speech template of the experimental customer service with the highest business completion rate in the Mth period as the optimal speech template; or, when the experimental customer service with the highest business completion rate is the same for N consecutive times, determine the speech template of the experimental customer service with the highest business completion rate as the optimal speech template; wherein N is a positive integer greater than 2.

[0031] In a third aspect, an embodiment of the present application further provides a computing device, including:

[0032] A memory for storing program instructions;

[0033] The processor is used to call the program instructions stored in the memory, and execute any method of implementing the above-mentioned first aspect according to the obtained program instructions.

[0034] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, in which computer-readable instructions are stored. When a computer reads and executes the computer-readable instructions, any method of the above-mentioned first aspect is implemented.

[0035] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising a computer program executable by a computer device, wherein when the program is run on the computer device, the computer device executes any method for implementing the above-mentioned first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flowchart of a method for determining a speech template provided in an embodiment of the present application;

[0037] Figure 2 A flowchart of a method for determining a speech template provided in an embodiment of the present application;

[0038] Figure 3 A flowchart of a method for determining a speech template provided in an embodiment of the present application;

[0039] Figure 4 A schematic diagram of the structure of a speech template system provided in an embodiment of the present application;

[0040] Figure 5 A schematic diagram of the structure of a device for determining a speech template provided in an embodiment of the present application;

[0041] Figure 6 A schematic diagram of the structure of a device for determining a speech template provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] Figure 1 A flowchart of a method for determining a speech template provided in an embodiment of the present application, the method can be executed by a speech template determination device, the speech template determination device can be a terminal device or a module for a terminal device, or a server or a module for a server. The present application does not limit the execution subject of the method.

[0043] The method comprises the following steps:

[0044] Step 101, assigning target customers with the same business attributes in the i-th period to multiple experimental customer services in the i-th period.

[0045] Among them, the business model of any experimental customer service is obtained by inputting each speech template into each generative large model for training; different experimental customer services have different business models; i is a positive integer.

[0046] In a possible implementation method, the business attributes include outbound collection, overdue reminders, or product marketing, etc., and this application does not limit the types of business attributes. For example, the business attribute is a 10-day overdue reminder, and the target customers with the same business attributes in the i-th cycle are assigned to multiple experimental customer services in the i-th cycle, including: in the i-th cycle, the target customers who have not repaid their loans for 10 days overdue are obtained, and the target customers who have not repaid their loans for 10 days overdue are assigned to multiple experimental customer services.

[0047] In one possible implementation method, target customers with the same business attributes are evenly distributed to multiple experimental customer service staff; or, target customers with the same business attributes are distributed according to the business processing capability of each experimental customer service staff. This application does not limit the specific distribution method.

[0048] In one possible implementation method, each experimental customer service is a machine customer service, each machine customer service corresponds to at least one business model, and different business models have different speech models. In this application, each experimental customer service corresponds to one business model as an example for explanation, which will not be repeated here.

[0049] In a possible implementation method, each business model is obtained by inputting each speech template into each generative big model for training. Specifically, the same speech template can be input into different generative big models for training to obtain different business models, for example, the collection speech template A is input into the generative big model A to obtain business model A, and the collection speech template A is input into the generative big model B to obtain business model B; different speech templates can also be input into the same generative big model for training to obtain different business models, for example, the collection speech template A1 is input into the generative big model A to obtain business model A1, and the collection speech template B1 is input into the generative big model A to obtain business model B1; different speech templates can also be input into different generative big models for training to obtain different business models, for example, the collection speech template A2 is input into the generative big model A to obtain business model A2, and the collection speech template B2 is input into the generative big model B to obtain business model B2; this application does not limit this. This application uses different types of generative big model training to obtain the business model, fully considering the differences between different generative big models, avoiding the limitations of single big model evaluation, and improving the accuracy and reliability of call quality evaluation.

[0050] In one possible implementation method, the generative large model is a model such as Transformer, GPT, and ChatGLM. This application does not limit the type of the generative large model.

[0051] Step 102, receiving the call contents of different experimental customer service staff in the i-th period.

[0052] The call content is obtained by different experimental customer services in the i-th period talking with corresponding target customers according to their own business models.

[0053] In a possible implementation method, after receiving the corresponding target customer, each experimental customer service will dial the target customer's phone number in sequence or randomly, conduct a call, and record the content of the call.

[0054] Step 103, based on the conversation contents of different experimental customer service staff in the i-th period, determine the speech template corresponding to the experimental customer service staff with the highest business completion rate in the i-th period.

[0055] In a possible implementation method, the call contents of different experimental customer services in the i-th period are quality inspected to determine the experimental customer service with the highest business completion rate in the i-th period. For example, the i-th period includes two experimental customer services, each of which corresponds to a business model. Experimental customer service A made collection calls to 1,000 target customers in the i-th period, of which 900 target customers repaid the debts, and the business completion rate of experimental customer service A in the i-th period was 90%; experimental customer service B made collection calls to 1,000 target customers in the i-th period, of which 500 target customers repaid the debts, and the business completion rate of experimental customer service B in the i-th period was 50%; then, the experimental customer service with the highest business completion rate in the i-th period is experimental customer service A.

[0056] Step 104, determine whether any customer service speech in the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period has a sub-speech.

[0057] Step 105: If it exists, expand the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the sub-speech.

[0058] In one possible implementation method, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is determined, and it is determined whether any customer service speech in the speech template has a sub-speech; if so, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded according to the sub-speech, and step 106 is executed; if not, based on the multiple business models of the i-th period, it is determined that multiple experimental customer services in the i+1th period enter the i+1th iteration process.

[0059] For example, the script template corresponding to the experimental customer service with the highest business completion rate in the i-th cycle is: The first customer service script: Hello, are you Mr. XXX? The customer's intention is: Yes; The second customer service script is: You have a loan that is overdue for 10 days, please repay it as soon as possible. The customer's intention is: No. The third customer service script is: OK. The experimental customer service executes the call hanging operation.

[0060] Analyze whether there are sub-scripts in the third customer service script when it is determined that the customer’s intention is not to return the money.

[0061] When it is determined that the customer's intention is not to repay, expand the first sub-speech to: you can choose to repay in installments; expand the second sub-speech to: you can choose the loan business; expand the third sub-speech to: you can borrow money.

[0062] In a possible implementation method, different tones are set for different expanded sub-speech, and the tones are divided into loose tone, ordinary tone and severe tone. Of course, other tones may also be included, and this application does not limit this.

[0063] Step 106, input the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the sub-speech into the generative large model for training to obtain the first business model.

[0064] In one possible implementation method, the speech template and different sub-speech corresponding to the experimental customer service with the highest business completion rate in the i-th period are input into the generative big model for training to obtain multiple first business models; illustratively, the speech template and the first sub-speech corresponding to the experimental customer service with the highest business completion rate in the i-th period are input into the generative big model for training to obtain the first business model A; the speech template and the second sub-speech corresponding to the experimental customer service with the highest business completion rate in the i-th period are input into the generative big model for training to obtain the first business model B; the speech template and the third sub-speech corresponding to the experimental customer service with the highest business completion rate in the i-th period are input into the generative big model for training to obtain the first business model C.

[0065] Step 107: Determine, based on the first business model and the multiple business models of the i-th period, that multiple experimental customer services of the i+1-th period enter the i+1-th iteration process.

[0066] The present application does not limit the specific time of each cycle, for example, each cycle is each day, or each cycle is each week, etc.

[0067] In one possible implementation method, the speech template of the experimental customer service with the highest business completion rate in the Mth cycle is determined as the optimal speech template; where M is the number of iteration terminations.

[0068] In a possible implementation method, when the experimental customer service with the highest business completion rate is the same for N consecutive times, the speech template of the experimental customer service with the highest business completion rate is determined as the optimal speech template; wherein N is a positive integer greater than 2.

[0069] In the above scheme, on the one hand, the business model of the experimental customer service is obtained by inputting each speech template into each generative large model for training. Compared with manual customer service, it is easier to expand the number of experimental customer service. Since each experimental customer service corresponds to at least one speech template, the speech template can be divided into smaller granularity, so that the business completion of different speech templates can be better compared, and faster iterative convergence can be achieved. However, the number and energy of manual customer service are limited. If the speech template is divided according to manual customer service, the granularity of the speech template is large. The business completion of the speech template may be good in a certain period, but the business completion of the speech template in the next period is poor, with large volatility and difficulty in convergence. On the other hand, the speech template corresponding to the experimental customer service with the highest business completion is divided into smaller granularity, which can achieve rapid convergence. On the other hand, the evaluation and expansion of the speech template in each period are obtained based on the actual communication content with the target customer, which can meet the needs of the target customer, thereby improving the business completion and user experience.

[0070] In a possible implementation method, before the above step 101, it further includes fine-tuning the generative large model according to the speech template to obtain different business models, wherein the speech template includes speech, tone, and intended response sequence. Exemplarily, the speech template is shown in Table 1.

[0071] Table 1

[0072]

[0073]

[0074] In a possible implementation method, before assigning target customers with the same business attributes to multiple experimental customer services in the above step 101, it also includes: grouping target customers according to the type of target customers, assigning each group of target customers to the corresponding multiple experimental customer services, and determining the speech template for each group of target customers. The types include but are not limited to age, gender, nationality, etc. Exemplarily, different speech templates need to be trained for different age groups, and the target customers are divided into a youth group, a middle-aged group, and an elderly group according to the age of the target customers, and the youth group is assigned to multiple experimental customer services corresponding to the youth group. The multiple experimental customer services corresponding to the youth group are mainly used to determine the speech template for the business attributes of the youth group; the middle-aged group is assigned to multiple experimental customer services corresponding to the middle-aged group, and the multiple experimental customer services corresponding to the middle-aged group are mainly used to determine the speech template for the business attributes of the middle-aged group; the elderly group is assigned to multiple experimental customer services corresponding to the elderly group, and the multiple experimental customer services corresponding to the elderly group are mainly used to determine the speech template for the business attributes of the elderly group.

[0075] In a possible implementation method, in the above step 101, each experimental customer service may include multiple business models, and different business models correspond to different speech templates.

[0076] In a possible implementation method, in the above step 101, each experimental customer service representative can dial the phone numbers of multiple target customers at the same time to improve work efficiency.

[0077] In another embodiment, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded as follows: Figure 2 As shown, the method comprises the following steps:

[0078] Step 201, determining whether there is a new intention in the call content of different experimental customer service staff in the i-th period.

[0079] A possible implementation method is to retrieve the call content of different experimental customer service personnel in the i-th period, and determine whether there is a new intention in the call content of different experimental customer service personnel in the i-th period. If so, expand the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the new intention. Exemplarily, the call content of a target customer in the i-th period includes a "complaint" label, but the different speech templates in the i-th period do not have "complaint" related speech, then according to the "complaint" label, expand the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period, including but not limited to; that is, for any speech template, the expansion operation can be performed according to the "complaint" label.

[0080] In a possible implementation method, the response order of different intents in the speech template is adjusted according to the new intent. For example, after determining the identity of the target customer, according to the second customer service speech, "You have a loan that is overdue for 10 days and you have not repaid it. Please repay it as soon as possible." However, the target customer replied, "Don't call me again. If you call again, I will complain about you." At this time, the experimental customer service modified the speech that was originally supposed to introduce the installment repayment business to the target customer, and responded to the "complaint" intent as soon as possible. Then the speech after the second customer service speech should be the speech for the complaint intent, such as "Sorry to bother you", and then respond to the speech introducing the installment repayment business.

[0081] Step 202: If it exists, expand the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the new intention.

[0082] Step 203, input the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the new intention into the generative big model for training to obtain the second business model.

[0083] In a possible implementation method, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded according to the new intent. For example, after determining that the new intent is "complaint", the first sub-speech is set to "Sorry, I'm bothering you, I won't bother you again in the future" for the "complaint" intent; the second sub-speech is "Complaint cannot solve the problem, please repay the money as soon as possible, otherwise you will bear the consequences yourself". Different tones are set for different sub-speech, for example, the tone of the first sub-speech is set to a relaxed tone, and the tone of the second sub-speech is set to a severe tone.

[0084] Step 204: Determine multiple experimental customer services for the (i+1)th period based on the first business model, the second business model, and the multiple business models for the i-th period.

[0085] In one possible implementation method, if only the first business model exists, then the multiple experimental customer services for the (i+1)th period are determined based on the first business model and the multiple business models of the (i)th period; if only the second business model exists, then the multiple experimental customer services for the (i+1)th period are determined based on the second business model and the multiple business models of the (i)th period; if both the first business model and the second business model exist, then the multiple experimental customer services for the (i+1)th period are determined based on the first business model, the second business model and the multiple business models of the (i)th period.

[0086] In one embodiment, in order to ensure the integrity of the speech template, try not to delete multiple business models in the i-th period, that is, add a new business model based on the multiple business models in the i-th period, because a certain business model may have the lowest business completion in the first period, but the highest business completion in the second period. If the experimental customer service with the lowest business completion in the i-th period is deleted rashly, this part of the speech may be lost. However, if the business completion of a certain experimental customer service is the lowest in multiple consecutive periods, the speech template corresponding to the experimental customer service can be deleted.

[0087] The above scheme, on the one hand, can divide the speech template into smaller granularities according to the new intent, so as to achieve better comparison of the business completion of different speech templates and achieve faster iterative convergence; on the other hand, each new intent is determined according to the specific call content, that is, the new intent is provided by the target customer and can reflect the real needs of the target customer, so that the expanded business template fits the needs of the target customer, thereby improving business completion and user experience.

[0088] In one embodiment, in the above step 103, according to the call contents of different experimental customer services in the i-th period, the speech template corresponding to the experimental customer service with the highest service completion rate in the i-th period is determined, such as Figure 3 As shown, the method comprises the following steps:

[0089] Step 301: Input the call contents of different experimental customer service staff in the i-th period into the quality inspection big model to obtain the quality inspection result of each experimental customer service staff.

[0090] In one possible implementation method, the generative big model is fine-tuned according to the quality inspection role, speech scoring rules, tone scoring rules, and intent scoring rules to determine the quality inspection big model; the quality inspection role is used to indicate that the role type of the generative big model is quality inspection. The quality inspection big model can be a fine-tuned Transformer, GPT, ChatGLM, and other models. This application does not limit the type of the quality inspection big model.

[0091] Of course, this application can also manually conduct quality inspection on the call content of different experimental customer service staff, but due to the low efficiency of manual inspection and the strong personal subjective consciousness of manual inspection, it is difficult to conduct quality inspection on the call content fairly.

[0092] Step 302: Determine the experimental customer service with the highest service completion rate in the i-th period according to the quality inspection results of each experimental customer service.

[0093] Step 303: Determine a speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the call content of the experimental customer service with the highest business completion rate in the i-th period.

[0094] In the above scheme, on the one hand, the quality inspection model can quickly and effectively evaluate the call content of different experimental customer service staff, especially when the call content is large, which is more efficient than manual evaluation; on the other hand, the quality inspection model can achieve a relatively accurate evaluation of the call content of different experimental customer service staff, reducing the influence of subjective factors in manual evaluation.

[0095] In a possible implementation method, the content of each call of the experimental customer service with the highest service completion rate in the i-th period is obtained; the content of each call is scored to obtain the scoring result of each call; the content of the call with the highest scoring result is determined as the speech template corresponding to the experimental customer service with the highest service completion rate in the i-th period. For example, the speech template used for training the experimental customer service with the highest service completion rate in the i-th period is shown in Table 2 below.

[0096] Table 2

[0097] First speech template Are you XXX? Customer intention: I Second speech template 1 You have a loan that is 10 days overdue. Please repay it as soon as possible. Customer Intention: Not Him Second speech template 2 Sorry to bother you. Goodbye.

[0098] Generative big models refer to generative models based on large-scale corpora, which are usually trained and generated in a manner similar to autoregressive models. Generative big models refer to large-scale neural network models that can generate, understand, and reason about natural language in an end-to-end manner. The technical principles of generative big models are mainly based on deep learning and natural language processing technologies. These models are usually trained and generated in an end-to-end manner, which can be understood as a mapping relationship between text input and output.

[0099] That is to say, even if the speech template in Table 2 is used as input to fine-tune the generative big model, the output of the generative big model may not be exactly the same as the speech template, and even if the business model trained by the same type of generative big model is obtained, the answers of the experimental customer service in each call content may not be exactly the same. For example, when the speech template in Table 2 is used to fine-tune the generative big model and business model A is obtained, during the call with customer A, business model A hits customer A's intention as "it's me", then business model A will be fine-tuned based on the second speech template 1 "You have a loan that is overdue for 10 days and has not been repaid. Please repay it as soon as possible", for example, the output speech is "Good afternoon, XXX, you have a loan that is overdue for 10 days and has not been repaid. Please repay it as soon as possible". In another call, during the conversation with customer B, business model A hits customer B's intention as "it's me", then business model A will make fine adjustments based on the second speech template 1 "You have a loan that is overdue for 10 days, please repay it as soon as possible", for example, the output speech is "You have a loan that is overdue for 10 days, please repay it within today". Business model A hits customer C's intention as "not me", then business model A will make fine adjustments based on the second speech template 2 "sorry, sorry to bother you, goodbye", for example, the output speech is "sorry, sorry to bother you, I wish you a happy life, goodbye".

[0100] After evaluation by the large quality inspection model, when the business model A outputs the speech "Good afternoon, XXX, you have a loan that is overdue for 10 days, please repay it as soon as possible", the call content score is the highest, that is, if the time tag is added, the target customer may feel better, then the speech output by the business model A at this time replaces the original second speech template 1 ("You have a loan that is overdue for 10 days, please repay it as soon as possible"), as the speech template corresponding to the experimental customer service with the highest business completion in the i-th cycle. This application only uses one round of conversation between the customer and the experimental customer service as an illustration. It is understandable that during each call, the target customer and the experimental customer service will have multiple rounds of conversations. The content of the call with the highest score is determined based on the comprehensive score of the multiple rounds of conversations as the speech template corresponding to the experimental customer service with the highest business completion in the i-th cycle.

[0101] In the above scheme, since the content of each call made by the experimental customer service may be different from the training speech template corresponding to the experimental customer service, the training speech template corresponding to the experimental customer service can be adjusted according to the content of the call with the highest score, so that the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th cycle is more in line with the actual application scenario, and further, it can improve business completion and user experience.

[0102] In one embodiment, the content of each call is scored to obtain the scoring result of each call, including: scoring the content of each call based on the hit rate of the words, the degree of matching between the words and the preset template, and / or the context relevance; scoring the tone of each call based on the rationality of the tone and / or the harmlessness of the tone; scoring the intention response of each call based on the rationality of the intended response sequence, the degree of matching between the intended response sequence and the preset template and / or the number of intended responses; and obtaining the scoring result of each call based on the words score, the tone score and the intended response score. Of course, the scoring result of each call can also be obtained based on one or more of the words score, the tone score and the intended response score, which is not limited in this application. This scheme can accurately and effectively determine the scoring result of each call based on the words score, the tone score and the intended response score.

[0103] In one possible implementation method, the hit rate of the speech script refers to the ratio of the number of hit speech script templates to the number of experimental customer service speech scripts in the content of each call. For example, in a call, the experimental customer service said a total of 10 sentences, but the number of hit speech script templates was only 5, then the hit rate of the speech script was 50%.

[0104] In a possible implementation method, the matching degree between the speech and the preset template refers to the similarity between the speech of the experimental customer service and the preset template. This application does not limit the method for determining the similarity.

[0105] In one possible implementation method, context relevance refers to the degree of relevance between each sentence of the experimental customer service and the context.

[0106] In one possible implementation method, the rationality of tone refers to whether the tone used in the current speech to the target customer is reasonable and the degree of influence of the current tone on the target customer. For example, if the experimental customer service currently uses a very harsh tone to answer the target customer's questions, causing the target customer to complain, or the target customer suddenly becomes very emotional, then the current tone of the experimental customer service is considered unreasonable.

[0107] In one possible implementation method, the harmlessness of tone refers to whether the experimental customer service has an aggressive or abusive tone, contains swear words or language that is detrimental to national interests, etc.

[0108] In one possible implementation method, the rationality of the intent response sequence refers to whether the intent response sequence of the experimental customer service is reasonable. For example, the target customer answers, "I can pay back the money today, but the bank is closed and I can't withdraw money. Can I pay it back tomorrow?" The intent response sequence should be "Can I pay it back tomorrow?" and "I can't withdraw money." At this time, the target customer should be clearly told that he cannot pay back the money tomorrow. If he cannot withdraw money, he can use operations such as transfer.

[0109] In one possible implementation method, the matching degree of the preset template specifies the order of responses to different intentions, and the matching degree of the intention response order of the experimental customer service can be determined with the preset template.

[0110] In one possible implementation method, the experimental customer service may ignore or fail to recognize the customer's intention, and the number of experimental customer service intention responses can be determined.

[0111] In another embodiment, in addition to the above-mentioned speech scoring rules, tone scoring rules and intention scoring rules, the quality inspection model can also output the scoring reasons and optimization points for each call content. Among them, the scoring reasons and optimization points can be for one or more of the above-mentioned speech scoring, tone scoring and intention scoring, and this application does not limit this.

[0112] In one possible implementation method, the speech template for the i-th cycle is optimized based on the scoring reasons and optimization points output by the large quality inspection model.

[0113] In one embodiment, the experimental customer service includes at least one manual customer service, and allocating the target customer of the i-th period to multiple experimental customer service personnel of the i-th period includes:

[0114] The target customer of the i-th period is allocated to at least two experimental customer service staff and at least one manual customer service staff of the i-th period; during the conversation with the target customer, the manual customer service staff can adaptively adjust the speech template according to the business attributes.

[0115] Human customer service has strong adaptability and can often give reasonable answers to target customers' words. At least one human customer service can be assigned as an experimental customer service to participate in the optimization process of the speech template. During the conversation with the target customer, the human customer service can adaptively adjust the speech template according to the business attributes. That is, if the business attribute is a collection business, the human customer service can fine-tune the speech template during the call without deviating from the collection business attribute.

[0116] In the early stages of iteration, the script template is often relatively simple. For example, there are only two business models, one with a relaxed tone and the other with a strict tone. The script template also has only two sentences, including confirming the identity of the target customer and informing the target customer that there is overdue business that needs to be processed. If manual customer service is used to participate in quality inspection, manual customer service is often the experimental customer service with the highest business completion rate. At this time, the script template can be quickly and effectively expanded based on the call content of the manual customer service, so that the iteration process can converge quickly.

[0117] In a possible implementation method, the present application provides a system for determining a speech template, such as Figure 4 As shown, it includes an experimental customer service module, a storage module and a quality inspection module. Among them, the experimental customer service module is used to execute calls; the storage module is used to store the call contents of different experimental customer services; the quality inspection module is used to perform quality inspection on the call contents of different experimental customer services, and determine the optimal speech template, and perform feedback optimization on the experimental customer service module. The experimental customer service module is also used to optimize the speech template and business model according to the feedback optimization of the quality inspection module. Among them, Figure 4 The business task agent refers to the business model.

[0118] Based on the same technical concept, Figure 5 The following is an exemplary diagram of a device 500 for determining a speech template provided by an embodiment of the present application. Figure 5 As shown, it includes: a transceiver unit 501 and a processing unit 502. The transceiver unit 501 is used to assign target customers with the same business attributes in the i-th period to multiple experimental customer services in the i-th period; the business model of any experimental customer service is obtained by inputting each speech template into each generative large model for training; wherein different experimental customer services have different business models; i is a positive integer; receiving the call content of different experimental customer services in the i-th period; the call content is obtained by the different experimental customer services in the i-th period talking with the corresponding target customers according to their own business models; the processing unit 502 is used to determine the business model of the i-th period according to the call content of different experimental customer services in the i-th period. The speech template corresponding to the experimental customer service with the highest degree of business completion; determine whether any customer service speech in the speech template corresponding to the experimental customer service with the highest degree of business completion in the i-th period has a sub-speech; if so, expand the speech template corresponding to the experimental customer service with the highest degree of business completion in the i-th period according to the sub-speech; input the speech template corresponding to the experimental customer service with the highest degree of business completion in the i-th period and the sub-speech into the generative large model for training to obtain a first business model; determine, based on the first business model and multiple business models of the i-th period, that multiple experimental customer services in the i+1th period enter the i+1th iteration process.

[0119] In a possible implementation method, the processing unit 502 is used to determine whether there is a new intention in the call content of different experimental customer service personnel in the i-th period; if so, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded according to the new intention; the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the new intention are input into the generative large model for training to obtain a second business model; and multiple experimental customer services in the (i+1)th period are determined according to the first business model, the second business model and multiple business models of the i-th period.

[0120] In one possible implementation method, the processing unit 502 is used to input the call contents of different experimental customer services in the i-th period into the quality inspection big model to obtain the quality inspection result of each experimental customer service; determine the experimental customer service with the highest business completion rate in the i-th period based on the quality inspection result of each experimental customer service; and determine the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period based on the call contents of the experimental customer service with the highest business completion rate in the i-th period.

[0121] In one possible implementation method, the processing unit 502 is used to obtain the call content of each call made by the experimental customer service with the highest business completion rate in the i-th period; score the call content of each call to obtain the scoring result of each call; and determine the call content of the call with the highest scoring result as the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period.

[0122] In one possible implementation method, the processing unit 502 is used to score the speech content of each call based on the hit rate of the speech, the degree of match between the speech and the preset template, and / or the context relevance; to score the tone of the call content of each call based on the rationality of the tone and / or the harmlessness of the tone; to score the intention response of the call content of each call based on the rationality of the intention response sequence, the degree of match between the intention response sequence and the preset template and / or the number of intention responses; and to obtain the scoring result for each call based on the speech score, the tone score and the intention response score.

[0123] In one possible implementation method, the processing unit 502 is used to fine-tune the generative big model according to the quality inspection role, speech scoring rules, tone scoring rules and intention scoring rules to determine the quality inspection big model; the quality inspection role is used to indicate that the role type of the generative big model is quality inspection.

[0124] In a possible implementation method, the transceiver unit 501 is used to assign the target customer of the i-th period to at least two experimental customer services and at least one manual customer service of the i-th period; during the conversation with the target customer, the manual customer service can adaptively adjust the speech template according to the business attributes.

[0125] In one possible implementation method, the processing unit 502 is used to determine the speech template of the experimental customer service with the highest business completion rate in the Mth period as the optimal speech template; or, when the experimental customer service with the highest business completion rate is the same for N consecutive times, determine the speech template of the experimental customer service with the highest business completion rate as the optimal speech template; wherein N is a positive integer greater than 2.

[0126] Based on the same technical concept, the embodiment of the present application provides a device 600 for determining a speech template, and the device 600 for determining a speech template may be, for example, a computing device. Figure 6 As shown, a device 600 for determining a speech template includes at least one processor 601 and a memory 602 connected to the at least one processor. In the embodiment of the present application, the specific connection medium between the processor 601 and the memory 602 is not limited. Figure 6 For example, the processor 601 and the memory 602 are connected via a bus. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0127] In an embodiment of the present application, the memory 602 stores instructions that can be executed by at least one processor 601. The at least one processor 601 can execute the above-mentioned method for determining a speech template by executing the instructions stored in the memory 602.

[0128] Among them, the processor 601 is a control center of a speech template determination device 600, which can use various interfaces and lines to connect various parts of the computer device, and perform resource settings by running or executing instructions stored in the memory 602 and calling data stored in the memory 602. Optionally, the processor 601 may include one or more determination units, and the processor 601 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 601. In some embodiments, the processor 601 and the memory 602 may be implemented on the same chip, and in some embodiments, they may also be implemented separately on independent chips.

[0129] Processor 601 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware processor for execution, or can be executed by a combination of hardware and software modules in the processor.

[0130] The memory 602 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory 602 may include at least one type of storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (Random Access Memory, RAM), a static random access memory (Static Random Access Memory, SRAM), a programmable read-only memory (Programmable Read Only Memory, PROM), a read-only memory (Read Only Memory, ROM), an electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), a magnetic memory, a disk, an optical disk, etc. The memory 602 is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 602 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0131] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer-executable program. The computer-executable program is used to enable a computer to execute a method for determining a speech template listed in any of the above methods.

[0132] An embodiment of the present application provides a computer program product, including a computer program executable by a computer device. When the program is run on the computer device, the computer device executes a method for determining a speech template listed in any of the above-mentioned methods.

[0133] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0134] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0135] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0137] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for determining a speech template. It is characterized in that The speech template is adjusted through M iterations to determine the optimal speech template, where M is a positive integer greater than 2. The i-th iteration process includes: Assign target customers with the same business attributes in the i-th period to multiple experimental customer services in the i-th period; the business model of any experimental customer service is obtained by inputting each speech template into each generative large model for training; different experimental customer services have different business models; i is a positive integer; receiving the call contents of different experimental customer services in the i-th period; the call contents are obtained by the different experimental customer services in the i-th period through calls with corresponding target customers according to their own business models; According to the conversation contents of different experimental customer service staff in the i-th period, determine the speech template corresponding to the experimental customer service staff with the highest business completion rate in the i-th period; Determine whether any customer service speech in the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period has a sub-speech; If it exists, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded according to the sub-speech; Input the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the sub-speech into the generative large model for training to obtain a first business model; According to the first business model and the multiple business models of the i-th period, it is determined that multiple experimental customer services of the i+1-th period enter the i+1-th iteration process.

2. The method according to claim 1, It is characterized in that The method further comprises: Determine whether there is a new intention in the call content of different experimental customer service in the i-th period; If so, the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is expanded according to the new intention; Input the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the new intention into the generative big model for training to obtain a second business model; The determining, according to the first business model and the multiple business models of the i-th period, multiple experimental customer services of the (i+1)th period includes: A plurality of experimental customer services in the (i+1)th period are determined according to the first business model, the second business model and the plurality of business models in the i-th period.

3. The method according to claim 1, It is characterized in that Determining the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the call contents of different experimental customer service personnel in the i-th period includes: Input the call contents of different experimental customer service staff in the i-th period into the quality inspection big model to obtain the quality inspection result of each experimental customer service staff; According to the quality inspection results of each experimental customer service, determine the experimental customer service with the highest business completion rate in the i-th cycle; According to the call content of the experimental customer service with the highest business completion rate in the i-th period, a speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period is determined.

4. The method according to claim 3, It is characterized in that Determining the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the call content of the experimental customer service with the highest business completion rate in the i-th period includes: Obtain the content of each call made by the experimental customer service representative with the highest business completion rate in the i-th period; Score the content of each call and obtain the score result of each call; The content of the call with the highest score is determined as the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th cycle.

5. The method according to claim 4, It is characterized in that Scoring the content of each call to obtain a scoring result for each call includes: The content of each call is scored based on the hit rate of the speech, the degree of match between the speech and the preset template, and / or the contextual relevance; The tone of each call was scored based on reasonableness of tone and / or innocence of tone; The intent response score of each call is assigned based on the rationality of the intent response sequence, the matching degree of the intent response sequence with the preset template, and / or the number of intent responses; The scoring result of each call is obtained according to the speech score, the tone score and the intended response score.

6. The method according to claim 3, It is characterized in that The method for determining the quality inspection large model includes: According to the quality inspection role, speech scoring rules, tone scoring rules and intention scoring rules, the generative big model is fine-tuned to determine the quality inspection big model; the quality inspection role is used to indicate that the role type of the generative big model is quality inspection.

7. The method according to claim 1, It is characterized in that The allocating the target customer of the i-th period to a plurality of experimental customer services of the i-th period includes: The target customer of the i-th period is allocated to at least two experimental customer service staff and at least one manual customer service staff of the i-th period; during the conversation with the target customer, the manual customer service staff can adaptively adjust the speech template according to the business attributes.

8. The method according to any one of claims 1 to 7, It is characterized in that The step of adjusting the speech template through M iterations to determine the optimal speech template includes: Determine the speech template of the experimental customer service with the highest business completion rate in the Mth cycle as the optimal speech template; or, If the experimental customer service with the highest business completion rate is the same for N consecutive times, the speech template of the experimental customer service with the highest business completion rate is determined as the optimal speech template; wherein N is a positive integer greater than 2.

9. A device for determining a speech template, It is characterized in that The speech template is adjusted through M iterations to determine the optimal speech template, where M is a positive integer greater than 2, and the device includes a transceiver unit and a processing unit; The transceiver unit is used to assign target customers with the same service attributes in the i-th period to multiple experimental customer services in the i-th period; The business model of any experimental customer service is obtained by inputting each speech template into each generative large model for training; wherein different experimental customer services have different business models; i is a positive integer; receiving the call content of different experimental customer services in the i-th period; the call content is obtained by the different experimental customer services in the i-th period talking with the corresponding target customers according to their own business models; The processing unit is used to determine the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the call contents of different experimental customer services in the i-th period; determine whether any customer service speech in the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period has a sub-speech; if so, expand the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period according to the sub-speech; input the speech template corresponding to the experimental customer service with the highest business completion rate in the i-th period and the sub-speech into the generative large model for training to obtain a first business model; and determine, according to the first business model and multiple business models of the i-th period, that multiple experimental customer services in the i+1th period enter the i+1th iteration process.

10. A computer program product, It is characterized in that The invention comprises a computer program executable by a computer device, and when the program is run on the computer device, the computer device is caused to execute the steps of the method according to any one of claims 1 to 8.

Citation Information

Cited By

  • Voice information generation method and device of intelligent robot, electronic equipment and medium

    CN119814926A