Pre-sales suggestion generation method, suggestion generation model training method and related equipment

Abstract

The invention discloses a pre-sales suggestion generation method, a suggestion generation model training method and related equipment, and aims to solve the problems that pre-sales suggestions generated in the prior art are lack of pertinence, cannot fit actual scenes and are difficult to provide effective support for implementation of sales strategies. The pre-sales suggestion generation method comprises the steps of obtaining dialogue content of a dialogue related to a to-be-marketed business between an agent and a customer; inputting the dialogue content into a suggestion generation model to obtain pre-sales suggestions corresponding to the to-be-marketed business; wherein when the suggestion generation model is trained, the sum of the minimum availability loss value and the key point coverage rate loss value is taken as a training target; the availability loss value is used for representing the availability difference between the actual pre-sales suggestion sample set and a generative pre-sales suggestion sample set generated by the suggestion generation model based on the dialogue content sample set; the key point coverage rate loss value is used for representing the coverage rate difference between the key points of the generative pre-sales suggestion sample set and the dialogue content sample set.

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and in particular to a pre-sales suggestion generation method, a suggestion generation model training method, and related equipment. Background Art

[0002] In a highly competitive market environment, maintaining and expanding customer relationships is crucial to the continued growth of a company's business. Especially in telephone or online sales scenarios, providing personalized pre-sales advice to customers can help sales staff effectively guide customers' decision-making process, thereby improving sales conversion rates. In the prior art, a general generative model, such as the Bidirectional and Auto-Regressive Transformers (BART) model, is usually used to generate pre-sales advice. However, when the prior art uses a general generative model to generate pre-sales advice, it usually pays too much attention to the content fluency and content format of the pre-sales advice, resulting in the generated pre-sales advice lacking pertinence and failing to fit the actual sales scenario, making it difficult to provide effective support for the implementation of sales strategies.

[0003] Based on this, how to generate pre-sales suggestions that can effectively support and guide the sales process is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0004] The embodiment of the present application provides a method for generating a pre-sales suggestion, which is used to solve the problem in the prior art that when generating pre-sales suggestions, too much attention is usually paid to the content fluency and content format of the pre-sales suggestions, resulting in the generated pre-sales suggestions lacking pertinence and failing to fit the actual sales scenario, thereby failing to provide effective support for the implementation of sales strategies.

[0005] Embodiments of the present application also provide a pre-sale suggestion generating device, an electronic device, a computer-readable storage medium, and a computer program product.

[0006] The present application embodiment adopts the following technical solutions: In a first aspect, the present application provides a method for generating a pre-sale suggestion, comprising: Obtain the conversation content between the agent and the customer related to the business to be marketed; Input the conversation content into the suggestion generation model to obtain pre-sales suggestions corresponding to the business to be marketed; When training the suggestion generation model, the training objective is to minimize the sum of the availability loss value and the key point coverage loss value. The usability loss value is used to characterize the usability difference between the actual pre-sales suggestion sample set and the generative pre-sales suggestion sample set generated by the suggestion generation model based on the conversation content sample set; The key point coverage loss value is used to characterize the difference between the coverage of the expected key points of pre-sales recommendations of the generated pre-sales recommendation sample set and the conversation content sample set.

[0007] Optionally, the availability loss value is determined in the following manner: Calculate a first usability score of the generated pre-sales suggestion sample set by using a preset usability function according to the conversation content sample set and the generated pre-sales suggestion sample set; Calculating a second usability score of the actual pre-sales suggestion sample set by using a preset usability function according to the conversation content sample set and the actual pre-sales suggestion sample set; An availability loss value is determined based on an availability difference between the first availability score and the second availability score.

[0008] Optionally, the key point coverage loss value is determined in the following way: Determine the key points of the sample set of dialogue content; According to the coverage difference between the keywords and the key points set of the generative pre-sales recommendation sample set, the key points coverage loss value is determined.

[0009] Optionally, before inputting the conversation content into the suggestion generation model to obtain the pre-sales suggestion corresponding to the business to be marketed, the method further includes: Obtaining a sample set of conversation contents and a sample set of actual pre-sales suggestions for the sample set of conversation contents; Inputting the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; Determine the usability differences between the sample set of generated pre-sales suggestions and the sample set of actual pre-sales suggestions, and the coverage differences between the key points of the sample set of generated pre-sales suggestions and the sample set of conversation content; The suggestion generation model is iteratively updated and trained based on the availability difference and coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and the iterative update training is stopped to obtain the trained suggestion generation model.

[0010] Optionally, obtain the conversation content between the agent and the customer related to the business to be marketed, including: Capture conversations between agents and customers related to the business to be marketed; Input the conversation into the conversation content extraction model to obtain the conversation content; The dialogue content extraction model is trained by using a mixed data set consisting of an information extraction instruction data set and a knowledge instruction data set of dialogue samples as model input, and dialogue content samples of dialogue samples as model output.

[0011] Optionally, a mixed dataset is constructed as follows: Determine the correlation threshold, information entropy and data ratio of the information extraction instruction data set and the knowledge instruction data set in the mixed data set; According to the correlation threshold, information entropy and data ratio, data are selected from the information extraction instruction data set and the knowledge instruction data set to form a mixed data set.

[0012] Optionally, the conversation content includes at least one of the following: The content of the client's objection to the business to be marketed; The reasons why the client did not renew the pending marketing business; Quotation comparison information for the business to be marketed.

[0013] In a second aspect, the present application provides a suggestion generation model training method, comprising: Obtaining a sample set of conversation contents and a sample set of actual pre-sales suggestions for the sample set of conversation contents; Inputting the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; Determine the usability differences between the sample set of generated pre-sales suggestions and the sample set of actual pre-sales suggestions, and the coverage differences between the key points of the sample set of generated pre-sales suggestions and the sample set of conversation content; The suggestion generation model is iteratively updated and trained based on the availability difference and coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and the iterative update training is stopped to obtain the trained suggestion generation model.

[0014] In a third aspect, the present application provides a pre-sale suggestion generating device, including an information acquisition module and a suggestion generating module, wherein: An information acquisition module is used to acquire the conversation content between the agent and the customer related to the business to be marketed; A suggestion generation module is used to input the conversation content into a suggestion generation model to obtain pre-sales suggestions corresponding to the business to be marketed; When training the suggestion generation model, the training objective is to minimize the sum of the availability loss value and the key point coverage loss value. The usability loss value is used to characterize the usability difference between the actual pre-sales suggestion sample set and the generative pre-sales suggestion sample set generated by the suggestion generation model based on the conversation content sample set; The key point coverage loss value is used to characterize the difference between the coverage of the expected key points of pre-sales recommendations of the generated pre-sales recommendation sample set and the conversation content sample set.

[0015] Optionally, the device further includes an availability loss value determination module, which is used to: Calculate a first usability score of the generated pre-sales suggestion sample set by using a preset usability function according to the conversation content sample set and the generated pre-sales suggestion sample set; Calculating a second usability score of the actual pre-sales suggestion sample set by using a preset usability function according to the conversation content sample set and the actual pre-sales suggestion sample set; An availability loss value is determined based on an availability difference between the first availability score and the second availability score.

[0016] Optionally, the device further includes a key point coverage loss value determination module, which is used to: Determine the key points of the sample set of dialogue content; According to the coverage difference between the keywords and the key points set of the generative pre-sales recommendation sample set, the key points coverage loss value is determined.

[0017] Optionally, the device is also used for: Before inputting the conversation content into the suggestion generation model to obtain the pre-sales suggestion corresponding to the business to be marketed, obtaining a conversation content sample set and an actual pre-sales suggestion sample set of the conversation content sample set; Inputting the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; Determine the usability differences between the sample set of generated pre-sales suggestions and the sample set of actual pre-sales suggestions, and the coverage differences between the key points of the sample set of generated pre-sales suggestions and the sample set of conversation content; The suggestion generation model is iteratively updated and trained based on the availability difference and coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and the iterative update training is stopped to obtain the trained suggestion generation model.

[0018] Optional, information acquisition module, used to: Capture conversations between agents and customers related to the business to be marketed; Input the conversation into the conversation content extraction model to obtain the conversation content; The dialogue content extraction model is trained by using a mixed data set consisting of an information extraction instruction data set and a knowledge instruction data set of dialogue samples as model input, and dialogue content samples of dialogue samples as model output.

[0019] Optionally, the device further comprises a mixed data set construction module, which is used to: Determine the correlation threshold, information entropy and data ratio of the information extraction instruction data set and the knowledge instruction data set in the mixed data set; According to the correlation threshold, information entropy and data ratio, data are selected from the information extraction instruction data set and the knowledge instruction data set to form a mixed data set.

[0020] Optionally, the conversation content includes at least one of the following: The content of the client's objection to the business to be marketed; The reasons why the client did not renew the pending marketing business; Quotation comparison information for the business to be marketed.

[0021] In a fourth aspect, the present application provides a suggestion generation model training device, comprising a sample acquisition module, a sample processing module, an information determination module and an iterative training module, wherein: A sample acquisition module, used to acquire a conversation content sample set and an actual pre-sales suggestion sample set of the conversation content sample set; A sample processing module, used to input the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; An information determination module, used to determine the usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and the coverage difference between the key points of the generated pre-sales suggestion sample set and the conversation content sample set; The iterative training module is used to iteratively update the suggestion generation model based on the availability difference and the coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and then stop the iterative update training to obtain the trained suggestion generation model.

[0022] In a fifth aspect, the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the pre-sales recommendation generating method or the recommendation generating model training method as described above are implemented.

[0023] In a sixth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements the above-mentioned pre-sales suggestion generating method, or the steps of the suggestion generating model training method.

[0024] In a seventh aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the pre-sales recommendation generating method or the recommendation generating model training method as described above.

[0025] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: Since the suggestion generation model provided in the embodiment of the present application is trained by a specific training method, the suggestion generation model will subsequently focus on the key point coverage between the generated pre-sales suggestions and the target conversation information in the process of generating pre-sales suggestions, reduce the situation where key points in actual sales scenarios are ignored, and then enable the pre-sales suggestions generated by the model to better cover the key points in the conversation, thereby ensuring that the generated pre-sales suggestions are more targeted and can provide effective support for the implementation of sales strategies.

[0026] On the other hand, in the process of generating pre-sales suggestions, the suggestion generation model will also pay attention to the availability loss between the generated pre-sales suggestions and the actual pre-sales suggestions. In this way, it can ensure that the pre-sales suggestions generated by the model are closer to the actual sales scenario, thereby providing effective support and guidance for the implementation of sales strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1a A schematic diagram of an implementation flow of a method for generating pre-sales suggestions provided in an embodiment of the present application; Figure 1b A schematic diagram of an implementation flow of a method for obtaining conversation content provided in an embodiment of the present application; Figure 2 A schematic diagram of an implementation flow of a method for training a suggestion generation model provided in an embodiment of the present application; Figure 3 A specific structural diagram of a pre-sale suggestion generating device is provided for an embodiment of the present application; Figure 4 A specific structural schematic diagram of a suggestion generation model training device is provided for an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0029] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0030] Example 1 In order to solve the problem in the prior art that when generating pre-sales suggestions, too much attention is usually paid to the content fluency and content format of the pre-sales suggestions, resulting in the generated pre-sales suggestions lacking pertinence and failing to fit the actual sales scenario, thus making it difficult to provide effective support for the implementation of sales strategies, the embodiment of the present application provides a method for generating pre-sales suggestions.

[0031] The execution subject of the method may be various types of computing devices, or an application or application (Application, APP) installed on the computing device. The computing device may be, for example, a user terminal such as a mobile phone, a tablet computer, a smart wearable device, or a server.

[0032] For ease of description, the present application embodiment takes the execution subject of the method as an example of a server to introduce the method. Those skilled in the art can understand that the present application embodiment takes the server as an example to introduce the method, which is only an exemplary description and does not limit the scope of protection of the claims corresponding to the present solution.

[0033] Specifically, the implementation process of the method provided in the embodiment of the present application is as follows: Figure 1a As shown, the following steps are included: Step 11, obtaining the conversation content between the agent and the customer related to the business to be marketed.

[0034] In the embodiment of the present application, the business to be marketed refers to the insurance business to be marketed / sold. For example, the insurance business may specifically be life insurance, health insurance, property insurance, agricultural insurance, automobile insurance, travel insurance, and the like.

[0035] It should be noted that the above-mentioned insurance business is merely an exemplary description of the embodiment of the present application and does not impose any limitation on the embodiment of the present application.

[0036] Optionally, in the embodiment of the present application, considering that the conversation content is obtained mainly for providing personalized pre-sales suggestions to customers, so as to help sales personnel effectively guide the customer's decision-making process, thereby improving the sales conversion rate, the conversation content may include at least one of the following: The content of the client's objection to the business to be marketed; The reasons why the client did not renew the pending marketing business; Quotation comparison information for the business to be marketed.

[0037] The objections of the customer to the marketing business include the doubts, concerns or objections of the customer to the products or services of the marketing business. For example, the objections may include price objections, product objections, service objections, demand objections, etc.

[0038] The reasons why customers do not renew their business, including the reasons why customers choose not to continue to purchase or use the products or services of the business after the service period of the business ends. For example, the reasons for non-renewal may include price reasons, quality reasons of the products or services of the business, demand changes, market changes, and customer experience.

[0039] Quotation comparison information for the business to be marketed, including quotations from different suppliers or service providers collected during the sales process when potential or existing customers are considering whether to purchase or renew the products or services of the business to be marketed, as well as comparative analysis of these quotations.

[0040] It should be noted that the above-mentioned objection contents and reasons for non-continuation are merely an exemplary description of the embodiments of the present application and do not impose any limitation on the embodiments of the present application.

[0041] Secondly, in the embodiment of the present application, considering that the conversation between the agent and the customer is usually conducted through telephone or online chat, and the conversation content or chat content generated through telephone or online chat can be recorded or recorded in real time, therefore, taking the conversation mode as a call as an example, when obtaining the conversation content of the conversation between the agent and the customer related to the business to be marketed, the conversation between the agent and the customer related to the business to be marketed (that is, the conversation content generated during the call) can be obtained first; then, the conversation content during the call is converted into text through voice recognition technology. Finally, through natural language processing technology, the call content is analyzed and identified based on the converted text and preset keywords and rules, so as to extract the conversation content, including the customer's objection content to the business to be marketed, the customer's reason for not continuing the business to be marketed, or the quotation comparison information of the business to be marketed. Among them, the preset keywords and rules are used to extract the conversation content such as the customer's objection content to the business to be marketed, the customer's reason for not continuing the business to be marketed, or the quotation comparison information of the business to be marketed from the conversation between the agent and the customer related to the business to be marketed.

[0042] Alternatively, taking online chatting as an example, when obtaining the conversation content of the conversation between the agent and the customer related to the business to be marketed, the conversation between the agent and the customer related to the business to be marketed (that is, the chat record) can be obtained first; then, through natural language processing technology, the call content is analyzed and identified based on the chat record and preset keywords and rules, so as to extract the conversation content.

[0043] Alternatively, in an optional implementation, in addition to processing through natural language processing technology and preset keywords and rules, the conversation content can also be obtained based on the conversation between the agent and the customer related to the business to be marketed through an information extraction model, such as a BERT-CRF model or a UIE model.

[0044] It should be noted that the above-mentioned method of obtaining the conversation content is only an exemplary description of the embodiment of the present application and does not impose any limitation on the embodiment of the present application.

[0045] Optionally, in an embodiment of the present application, considering that in actual applications, when existing information extraction models (such as BERT-CRF model or UIE model, etc.) are used to obtain conversation content, these models themselves have some significant defects, which will affect the accuracy and completeness of the extracted conversation content. For example, the BERT-CRF model is limited by the length of the input data, which may lead to incomplete information extraction; and the UIE model performs poorly in processing long texts, lacks domain knowledge reasoning capabilities, and cannot accurately extract entity relationships. These problems will negatively affect the generation of subsequent pre-sales recommendations. Based on this, in order to avoid this problem, in an optional implementation, such as Figure 1b As shown, the conversation content of the conversation between the agent and the customer related to the business to be marketed can be obtained by using the pre-trained conversation content extraction model in the manner shown in the following steps 111 to 112: Step 111, obtaining the conversation between the agent and the customer related to the business to be marketed.

[0046] In an embodiment of the present application, when obtaining a conversation between an agent and a customer related to the business to be marketed, the conversation mode between the agent and the customer may be determined first, and then the conversation between the agent and the customer related to the business to be marketed may be obtained based on the conversation mode.

[0047] For example, taking the conversation mode as a phone call, when obtaining the conversation content of the conversation between the agent and the customer related to the business to be marketed, the conversation content related to the business to be marketed generated by the agent and the customer during the call can be obtained as the conversation between the agent and the customer related to the business to be marketed.

[0048] Alternatively, assuming that the conversation method is an online chat method, when obtaining the conversation content of the conversation between the agent and the customer related to the business to be marketed, the chat record related to the business to be marketed generated by the agent and the customer during the online chat can be obtained as the conversation between the agent and the customer related to the business to be marketed.

[0049] It should be noted that the above-mentioned method of obtaining the conversation between the seat and the customer related to the marketing business is only an exemplary description in the embodiment of the present application and does not impose any limitation on the embodiment of the present application.

[0050] Step 112, input the conversation into the conversation content extraction model to obtain the conversation content.

[0051] The dialogue content extraction model is trained by using a mixed data set consisting of an information extraction instruction data set and a knowledge instruction data set of dialogue samples as model input, and dialogue content samples of dialogue samples as model output.

[0052] The information extraction instruction dataset is used to train the conversation content extraction model, so that the model can extract the conversation content related to the business to be marketed from the conversation between the agent and the customer, including the customer's objection to the business to be marketed, the customer's reason for not renewing the business to be marketed, and the quotation comparison information of the business to be marketed.

[0053] In the embodiment of the present application, it is considered that when the dialogue content extraction model is applied to a specific field, the knowledge in the field may be unique, which may cause the dialogue content extraction model to be difficult to accurately capture the required information from the dialogue. Therefore, in order to avoid the above problems and ensure that the dialogue content extraction model can also accurately extract the required information in a specific technical field, in the embodiment of the present application, in addition to using the information extraction instruction data set, the dialogue content extraction model can be further combined with the professional knowledge in the specific field (that is, the knowledge instruction data set) to train the dialogue content extraction model, so as to ensure the accuracy and effectiveness of the information extraction of the trained model in the specific field.

[0054] The so-called knowledge instruction dataset can be understood as a dataset consisting of professional knowledge in a specific field, which is used to help the conversation content extraction model extract enhanced information related to a specific knowledge field from conversations in a specific field. For example, assuming that the business to be marketed is auto insurance business, the dataset consisting of professional knowledge in the auto insurance business field can be called a knowledge instruction dataset. Among them, the professional knowledge in the auto insurance business field includes, for example, the definition of auto insurance terms, the understanding of auto insurance terms, and auto insurance quotation rules.

[0055] Optionally, in an embodiment of the present application, before inputting the conversation into the conversation content extraction model to obtain the conversation content, it is necessary to first train the conversation content extraction model to obtain a trained conversation content extraction model.

[0056] In an optional implementation, the conversation content extraction model may be trained in the manner described in the following steps (1) to (3): (1) Construct a dataset of information extraction instructions for dialogue samples.

[0057] In the embodiment of the present application, taking the insurance business as an example, when constructing the information extraction instruction data set of the dialogue sample, the dialogue sample related to the insurance business between the agent and the customer can be obtained first, and then the dialogue sample can be preprocessed, for example, the "opening remarks" and "closing remarks" in the dialogue sample are removed. Meaningless and irrelevant information. After preprocessing, the dialogue sample is segmented based on keywords such as "quotation", "insurance type", and "amount type" of the insurance business to obtain the dialogue sample segmentation result. Finally, the quotation sheet matching the dialogue sample is obtained according to the semantic summary information of the dialogue sample, and the quotation information and the dialogue sample segmentation result are extracted using a regular method according to the quotation sheet to form an information extraction instruction data set.

[0058] The quotation information at least includes the quotation triplet (insurance type, insured amount, quotation) and the total price of the insurance business. It should be noted that the three entities of insurance type, insured amount and quotation in the quotation triplet do not need to exist at the same time, but there must be an "insurance type" entity.

[0059] In an optional implementation, in order to increase the diversity of the information extraction instruction data set, different descriptions of the same insurance type by agents in a real environment can be simulated, and the frequency of occurrence of various insurance types and insured amounts can be counted; then, based on the frequency, the EDA data enhancement method is adopted to construct a diverse information extraction instruction data set by random replacement, random addition, entity injection, etc.

[0060] For example, taking the random replacement method as an example, assuming that in actual applications, when describing "non-medical insurance expense liability insurance", the agent usually says "non-medical insurance expense liability insurance" as "non-medical insurance" or "non-medical insurance medication", that is, "non-medical insurance" or "non-medical insurance medication" appear more frequently in actual applications than "non-medical insurance expense liability insurance". In this case, in order to increase the diversity of the information extraction instruction data set, "non-medical insurance expense liability insurance" can be replaced with "non-medical insurance" or "non-medical insurance medication".

[0061] Secondly, the so-called entity injection, that is, for insurance types with smaller sample sizes, supplementary information is added in terms of quotes and amounts, thereby increasing the diversity of the information extraction instruction data set.

[0062] In addition, considering that the sample size of customer quotation feedback information in most data is low, in order to increase the diversity of the information extraction instruction data set, quotation triplets of quotation feedback can also be added to the customer information to ensure sample balance and increase the diversity of the information extraction instruction data set.

[0063] (2) Construct a knowledge instruction dataset.

[0064] Continuing with the above example, taking the insurance business as an example, in order to enable the dialogue content extraction model to better understand the insurance terminology, quotation rules, customer needs, etc. in the insurance business field, when constructing a knowledge instruction data set, the embodiment of the present application can first design a set of information extraction tasks related to the insurance business field; then formulate specific instructions for each information extraction task; finally, based on the instructions, construct insurance term definitions, insurance entity disambiguation, insurance clause understanding, and quotation compliance judgment through different knowledge enhancement strategies.

[0065] Among them, specific instructions are formulated for each information extraction task, mainly to guide the dialogue content extraction model on how to extract and enhance knowledge from the input information.

[0066] In the embodiment of the present application, when formulating specific instructions for each information extraction task, for example, for the insurance term definition task, the instruction can be set as: "Please explain the coverage of XXXX". For example, assuming that you need to understand the definition of medical insurance external medicines, the instruction can be specifically set as: "Please explain the coverage of medical insurance external medicines".

[0067] For the insurance entity disambiguation task, the instruction can be set as: "Please list the various common names of XXXX". Continuing with the above example, assuming that medical insurance external medicines need to be disambiguated, the instruction can be specifically set as: "Please list the various common names of medical insurance external medicines".

[0068] In summary, the big model can be used to transform the triples of insurance terms and regulations and insurance knowledge base into the form of "input question + task instruction: output response" according to different task types. The specific implementation method can be to input a task instruction, then extract the answer from the knowledge base, and finally obtain an insurance knowledge construction knowledge instruction dataset.

[0069] (3) A hybrid dataset is constructed based on the information extraction instruction dataset and the knowledge instruction dataset, and a dialogue content extraction model is trained using the hybrid dataset as the model input and the dialogue content samples of the dialogue samples as the model output.

[0070] In an embodiment of the present application, after obtaining the information extraction instruction data set and the knowledge instruction data set, the entire information extraction instruction data set and the entire knowledge instruction data set can be directly used as a mixed data set; then, the mixed data set is used as the model input, and the dialogue content sample of the dialogue sample is used as the model output to train the dialogue content extraction model.

[0071] Optionally, considering that the ratio of the information extraction instruction dataset to the knowledge instruction dataset in the mixed dataset will affect the stability and tendency of the dialogue content extraction model, in order to ensure the stability and tendency of the dialogue content extraction model, before constructing the mixed dataset, the knowledge distribution of the mixed dataset can be determined first, and then the mixed dataset can be determined based on the knowledge distribution.

[0072] Among them, when determining the knowledge distribution of a mixed data set, the following three regression features can be considered: the correlation threshold of the mixed data set, the information entropy, and the data ratio of the information extraction instruction data set and the knowledge instruction data set in the mixed data set.

[0073] In an optional implementation, the mixed data set may be constructed in the following manner: Firstly, the correlation threshold, information entropy and the data ratio of the information extraction instruction data set and the knowledge instruction data set in the mixed data set are determined.

[0074] Secondly, according to the relevance threshold, information entropy and data ratio, data are selected from the information extraction instruction data set and the knowledge instruction data set to form a mixed data set.

[0075] The correlation threshold of the mixed data set can be used to characterize the similarity between the information extraction instruction data set and the knowledge instruction data set in the mixed data set. In the embodiment of the present application, the correlation threshold of the mixed data set can be determined according to a preset perplexity threshold.

[0076] Information entropy can be used to characterize the degree of diversity of a mixed data set. In the embodiment of the present application, when calculating information entropy, the expected answers corresponding to the information extraction instruction data set and the knowledge instruction data set can be first obtained; then, the knowledge type of the expected answer is divided into different labels according to the task type, and the information entropy is calculated by the following calculation method:

[0077] in, represents information entropy; The label indicating the knowledge type of the expected answer is The probability of n is the number of labels.

[0078] Optionally, in the embodiment of the present application, when determining the data ratio of the information extraction instruction data set and the knowledge instruction data set in the mixed data set, the following method can be used: (1) Define input features: Determine the input features X = [x1, x2, x3], where x1 represents the data ratio of the information extraction instruction data set S1 and the knowledge instruction data set S2 in the combined data set; x2 represents the correlation threshold of the mixed data set; x3 represents the information entropy; (2) For each X, fix the information extraction instruction dataset S1, then select S1 / x1 samples with a perplexity less than x2 from the knowledge instruction dataset S2, and ensure that the information entropy of the keywords in the constructed mixed dataset is less than x3; (3) According to the constraints in step (2) above, determine the value range of x1; (4) According to the value range of x1, the correlation threshold, and the information entropy, data is selected from the information extraction instruction data set and the knowledge instruction data set to form the mixed data set S3.

[0079] In an optional implementation, in order to determine the optimal data ratio, in the embodiment of the present application, after executing the above step (4) to obtain the mixed data set S3, the following steps may be further performed: (5) Fine-tune S3 on a model with smaller parameters (such as the 1B model) to obtain the validation set results.

[0080] (6) Construct data labels: Use the validation set results obtained through fine-tuning as labels to form a piece of data.

[0081] (7) Afterwards, according to the value range of x1, different data ratios are used for fine-tuning multiple times to obtain multiple data.

[0082] (8) Fit the multiple data obtained in (7) through a linear regression function to obtain a prediction function.

[0083] (9) Based on the prediction function, predict the expected results of the validation set under different data ratios to determine the optimal data ratio.

[0084] After obtaining the optimal data ratio, data can be selected from the information extraction instruction data set and the knowledge instruction data set to form the optimal mixed data set based on the correlation threshold, information entropy and the optimal data ratio.

[0085] Step 12: Input the conversation content into a suggestion generation model to obtain pre-sales suggestions corresponding to the business to be marketed.

[0086] When training the suggestion generation model, the training objective is to minimize the sum of the availability loss value and the key point coverage loss value.

[0087] The usability loss value is used to characterize the usability difference between the actual pre-sales recommendation sample set and the generated pre-sales recommendation sample set generated by the recommendation generation model based on the conversation content sample set.

[0088] The key point coverage loss value is used to characterize the difference between the coverage of the expected key points of pre-sales recommendations of the generated pre-sales recommendation sample set and the conversation content sample set.

[0089] In an optional implementation, when determining the availability loss value, a first availability score of the generated pre-sales suggestion sample set may be calculated based on the conversation content sample set and the generated pre-sales suggestion sample set by a preset availability function; and a second availability score of the actual pre-sales suggestion sample set may be calculated based on the conversation content sample set and the actual pre-sales suggestion sample set by a preset availability function; and then, based on the availability difference between the first availability score and the second availability score, the availability loss value may be determined.

[0090] Optionally, the preset availability function in this application can be designed as follows:

[0091] Wherein, Y represents the availability score obtained based on the preset availability function. The concatenation result of the vector representation of the generated pre-sales suggestion sample set / the actual pre-sales suggestion sample set and the vector representation of the conversation content; Represents a preset availability function.

[0092] For example, to calculate the first usability score for a sample set of generative pre-sales recommendations , we can first determine the vector representation of the generative pre-sales suggestion sample set and the vector representation of the conversation content; then, based on the vector representation of the generative pre-sales suggestion sample set and the vector representation of the conversation content, determine the concatenation result of the two vector representations ; After getting the splicing result represented by these two vectors After that, the first availability score can be calculated as follows: :

[0093] Correspondingly, if we want to calculate the second usability score of the actual pre-sales recommendation sample set , we can first determine the vector representation of the actual pre-sales recommendation sample set and the vector representation of the conversation content; then, based on the vector representation of the actual pre-sales recommendation sample set and the vector representation of the conversation content, determine the concatenation result of the two vector representations. ; After getting the splicing result represented by these two vectors After that, the second availability score can be calculated as follows: :

[0094] Optionally, the availability loss value in the embodiment of the present application can be calculated by the following calculation method:

[0095] in, Indicates the availability loss value; A weight coefficient representing the difference in availability of presets.

[0096] In an optional implementation, when determining the key point coverage loss value, the key point set of the conversation content sample set can be determined first; then, the key point coverage loss value is determined based on the coverage difference between the keywords of the generative pre-sales recommendation sample set and the key point set.

[0097] In an embodiment of the present application, assuming that the key point set of the conversation content sample set is K, and each key point is k, when the similarity between the keywords in the generative pre-sales recommendation sample set and the key point k in the key point set is greater than or equal to the preset similarity threshold, it can be considered that the key point k is successfully covered; correspondingly, the loss is 0.

[0098] On the contrary, if the similarity between the keywords in the generative pre-sales recommendation sample set and the key point k in the key point set is less than the preset similarity threshold, it can be considered that the key point k is not covered; correspondingly, the loss is 1. Based on this, the key point coverage loss value in the embodiment of the present application can be determined by the following calculation method:

[0099] in, represents the key point coverage loss value; K represents the key point set; k represents the key point; represents the keyword set in the generative pre-sale suggestion sample set. t represents the keyword in the generative pre-sale suggestion sample set. represents the similarity between keyword t in the generative pre-sales recommendation sample set and key point k in the key point set. Indicates the preset similarity threshold.

[0100] In the embodiment of the present application, the recommendation generation model is trained using the key point coverage loss value calculated by the above calculation method, which can ensure the reliability of the generative pre-sales recommendations generated based on the recommendation generation model.

[0101] In an optional implementation, before inputting the conversation content into the suggestion generation model to obtain the pre-sales suggestion corresponding to the business to be marketed, the suggestion generation model may be trained to obtain a trained suggestion generation model.

[0102] Optionally, in the embodiment of the present application, the suggestion generation model may be trained in the following manner: (1) Obtain a dialogue content sample set and an actual pre-sales recommendation sample set of the dialogue content sample set.

[0103] (2) Input the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set.

[0104] (3) Determine the usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, as well as the coverage difference between the key points of the generated pre-sales suggestion sample set and the conversation content sample set.

[0105] Among them, the availability difference between the generated pre-sales recommendation sample set and the actual pre-sales recommendation sample set is the availability loss value.

[0106] Secondly, the coverage difference between the key points of the generative pre-sales recommendation sample set and the conversation content sample set, that is, the key point coverage loss value.

[0107] (4) The suggestion generation model is iteratively updated and trained based on the availability difference and coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and the iterative update training is stopped to obtain the trained suggestion generation model.

[0108] Among them, the suggestion generation model is iteratively updated and trained based on the availability difference and the coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, that is, the suggestion generation model is iteratively updated and trained based on the availability loss value and the key point coverage loss value until the sum of the availability loss value and the key point coverage loss value of the suggestion generation model after the iterative update training meets the preset minimization requirements.

[0109] In an optional implementation, when the suggestion generation model is iteratively updated and trained based on the availability loss value and the key point coverage loss value, the joint loss function value of the suggestion generation model can be first calculated based on the availability loss value and the key point coverage loss value according to the following calculation method:

[0110] in, Represents the joint loss function value of the proposed generation model; Represents the weight coefficient of the proposed generation model’s own loss; represents the self-loss of the proposed generation model; The weight coefficient representing the loss value of key point coverage; Indicates the coverage loss value of key points; A weight coefficient representing the value of the availability loss; Indicates the availability loss value.

[0111] Optional, weight factor for key point coverage loss and the weight coefficient of the availability loss value , can be determined as follows:

[0112] in, Indicates the average number of key points in the conversation content; Indicates the self-sustaining success rate of the business to be marketed; represents the total number of samples used to train the proposal generation model; Indicates compliance The number of samples.

[0113]

[0114] in, represents the average usability score of the sample set of generative pre-sales recommendations; Indicates compliance The number of samples.

[0115] The specific method of generating pre-sales suggestions in this application is described in detail below with reference to examples.

[0116] Specifically, for a chat process between an agent and a customer, first, we can retrieve three past dialogues between the agent and the customer (these three dialogues are related to the business to be marketed): “09-14”: “Agent: Hello.......Customer: Please help me change the three to 2 million…”; “09-15”: “Agent: Hello.......Customer: Insurance company A offers similar insurance to yours, but it’s cheaper…”; “09-16”: “Agent: We have driving insurance... Customer: Let me think about it...”

[0117] Secondly, based on the dialogues, we extract the objection questions, quotation sheets and other information, and combine them into an input abstracts:[“09-14”:“Objection question: the third party insurance is adjusted to 2 million; quotation sheet comparison: our company: third party insurance 3 million, compulsory traffic insurance, driving insurance”,“09-15”:“Objection question: Ping An is cheaper; quotation sheet: Insurance company A and our company have the same price, 300 less”,“09-16”:“Objection question: reconsider; quotation sheet comparison: none”] After that, the abstracts are input into the trained suggestion generation model to generate pre-sales suggestions: [“Overall evaluation: users are more sensitive to price and say they don’t need very high protection. Meanwhile, competitors’ prices are more attractive, but in the end, customers are swayed. It is recommended to focus on cost-effectiveness. The quote is more expensive: PICC’s price is 300 higher, but it has an additional driver and passenger insurance, emphasizing the cost-effectiveness of the driver and passenger insurance; if you consider it again, there is no rush: highlight that the insurance is about to expire and the current price is the best”].

[0118] When the method provided by the embodiment of the present application is adopted, since the suggestion generation model provided by the embodiment of the present application is obtained by training through a specific training method, the suggestion generation model will subsequently focus on the key point coverage between the generated pre-sales suggestions and the target dialogue information in the process of generating pre-sales suggestions, reduce the situation where key points in actual sales scenarios are ignored, and thus enable the pre-sales suggestions generated by the model to better cover the key points in the dialogue, thereby ensuring that the generated pre-sales suggestions are more targeted and can provide effective support for the implementation of sales strategies.

[0119] On the other hand, in the process of generating pre-sales suggestions, the suggestion generation model will also pay attention to the availability loss between the generated pre-sales suggestions and the actual pre-sales suggestions. In this way, it can ensure that the pre-sales suggestions generated by the model are closer to the actual sales scenario, thereby providing effective support and guidance for the implementation of sales strategies.

[0120] Example 2 In order to solve the problem that the existing general generative models usually pay too much attention to the content fluency and content format of pre-sales suggestions when generating pre-sales suggestions, resulting in the generated pre-sales suggestions lacking pertinence and failing to fit the actual sales scenarios, thus making it difficult to provide effective support for the implementation of sales strategies, the embodiment of the present application provides a suggestion generation model training method.

[0121] The execution subject of the method may be various types of computing devices, or an application or application (Application, APP) installed on the computing device. The computing device may be, for example, a user terminal such as a mobile phone, a tablet computer, a smart wearable device, or a server.

[0122] For ease of description, the present application embodiment takes the execution subject of the method as an example of a server to introduce the method. Those skilled in the art can understand that the present application embodiment takes the server as an example to introduce the method, which is only an exemplary description and does not limit the scope of protection of the claims corresponding to the present solution.

[0123] Specifically, the implementation process of the method provided in the embodiment of the present application is as follows: Figure 2 As shown, the following steps are included: Step 21, obtaining a dialogue content sample set and an actual pre-sales suggestion sample set of the dialogue content sample set.

[0124] Step 22: input the conversation content sample set into a suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set.

[0125] In an optional implementation, before inputting the conversation content sample set into the suggestion generation model to obtain the generative pre-sale suggestion sample set of the conversation content sample set, the suggestion generation model may be trained to obtain a trained suggestion generation model.

[0126] Among them, the specific method of training the suggestion generation model can refer to the relevant content in Example 1, which will not be repeated here.

[0127] Step 23, determining the usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and the coverage difference between the key points of the generated pre-sales suggestion sample set and the conversation content sample set.

[0128] Among them, the availability difference between the generated pre-sales recommendation sample set and the actual pre-sales recommendation sample set is the availability loss value.

[0129] Secondly, the coverage difference between the key points of the generative pre-sales recommendation sample set and the conversation content sample set, that is, the key point coverage loss value.

[0130] In an optional implementation, when determining the availability loss value, a first availability score of the generated pre-sales suggestion sample set may be calculated based on the conversation content sample set and the generated pre-sales suggestion sample set by a preset availability function; and a second availability score of the actual pre-sales suggestion sample set may be calculated based on the conversation content sample set and the actual pre-sales suggestion sample set by a preset availability function; and then, based on the availability difference between the first availability score and the second availability score, the availability loss value may be determined.

[0131] In an optional implementation, when determining the key point coverage loss value, the key point set of the conversation content sample set can be determined first; then, the key point coverage loss value is determined based on the coverage difference between the keywords of the generative pre-sales recommendation sample set and the key point set.

[0132] Step 24, iteratively update the suggestion generation model based on the availability difference and coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and stop the iterative update training to obtain the trained suggestion generation model.

[0133] By using the method provided in the embodiment of the present application, the suggestion generation model can be trained based on the usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and the coverage difference between the key points of the generated pre-sales suggestion sample set and the conversation content sample set. In this way, it can be ensured that the trained suggestion generation model can focus on the key point coverage between the generated pre-sales suggestion and the target conversation information, reduce the situation where the key points in the actual sales scenario are ignored, and then the pre-sales suggestions generated by the model can better cover the key points in the conversation, thereby ensuring that the generated pre-sales suggestions are more targeted and can provide effective support for the implementation of the sales strategy.

[0134] On the other hand, it can also ensure that the trained suggestion generation model can focus on the availability loss between the generated pre-sales suggestions and the actual pre-sales suggestions. In this way, it can ensure that the pre-sales suggestions generated by the model are closer to the actual sales scenario, thereby providing effective support and guidance for the implementation of sales strategies.

[0135] Example 3 In order to solve the problem in the prior art that when generating pre-sales suggestions, too much attention is usually paid to the content fluency and content format of the pre-sales suggestions, resulting in the generated pre-sales suggestions lacking pertinence and failing to fit the actual sales scenario, thereby failing to provide effective support for the implementation of sales strategies, the present application embodiment provides a pre-sales suggestion generating device, the specific structural diagram of the device is shown in FIG. Figure 3 As shown, it includes an information acquisition module 31 and a suggestion generation module 32. The functions of each module are as follows: The information acquisition module 31 is used to acquire the conversation content between the agent and the customer related to the business to be marketed; A suggestion generation module 32, used to input the conversation content into a suggestion generation model to obtain pre-sales suggestions corresponding to the business to be marketed; When training the suggestion generation model, the training objective is to minimize the sum of the availability loss value and the key point coverage loss value. The usability loss value is used to characterize the usability difference between the actual pre-sales suggestion sample set and the generative pre-sales suggestion sample set generated by the suggestion generation model based on the conversation content sample set; The key point coverage loss value is used to characterize the difference between the coverage of the expected key points of pre-sales recommendations of the generated pre-sales recommendation sample set and the conversation content sample set.

[0136] Optionally, the device further includes an availability loss value determination module, which is used to: Calculate a first usability score of the generated pre-sales suggestion sample set by using a preset usability function according to the conversation content sample set and the generated pre-sales suggestion sample set; Calculating a second usability score of the actual pre-sales suggestion sample set by using a preset usability function according to the conversation content sample set and the actual pre-sales suggestion sample set; An availability loss value is determined based on an availability difference between the first availability score and the second availability score.

[0137] Optionally, the device further includes a key point coverage loss value determination module, which is used to: Determine the key points of the sample set of dialogue content; According to the coverage difference between the keywords and the key points set of the generative pre-sales recommendation sample set, the key points coverage loss value is determined.

[0138] Optionally, the device is also used for: Before inputting the conversation content into the suggestion generation model to obtain the pre-sales suggestion corresponding to the business to be marketed, obtaining a conversation content sample set and an actual pre-sales suggestion sample set of the conversation content sample set; Inputting the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; Determine the usability differences between the sample set of generated pre-sales suggestions and the sample set of actual pre-sales suggestions, and the coverage differences between the key points of the sample set of generated pre-sales suggestions and the sample set of conversation content; The suggestion generation model is iteratively updated and trained based on the availability difference and coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and the iterative update training is stopped to obtain the trained suggestion generation model.

[0139] Optionally, the information acquisition module 31 is used to: Capture conversations between agents and customers related to the business to be marketed; Input the conversation into the conversation content extraction model to obtain the conversation content; The dialogue content extraction model is trained by using a mixed data set consisting of an information extraction instruction data set and a knowledge instruction data set of dialogue samples as model input, and dialogue content samples of dialogue samples as model output.

[0140] Optionally, the device further comprises a mixed data set construction module, which is used to: Determine the correlation threshold, information entropy and data ratio of the information extraction instruction data set and the knowledge instruction data set in the mixed data set; According to the correlation threshold, information entropy and data ratio, data are selected from the information extraction instruction data set and the knowledge instruction data set to form a mixed data set.

[0141] Optionally, the conversation content includes at least one of the following: The content of the client's objection to the business to be marketed; The reasons why the client did not renew the pending marketing business; Quotation comparison information for the business to be marketed.

[0142] When the device provided by the embodiment of the present application is used, since the suggestion generation model provided by the embodiment of the present application is obtained through training in a specific training method, the suggestion generation model will subsequently focus on the key point coverage between the generated pre-sales suggestions and the target dialogue information in the process of generating pre-sales suggestions, reduce the situation where key points in actual sales scenarios are ignored, and then enable the pre-sales suggestions generated by the model to better cover the key points in the dialogue, thereby ensuring that the generated pre-sales suggestions are more targeted and can provide effective support for the implementation of sales strategies.

[0143] On the other hand, in the process of generating pre-sales suggestions, the suggestion generation model will also pay attention to the availability loss between the generated pre-sales suggestions and the actual pre-sales suggestions. In this way, it can ensure that the pre-sales suggestions generated by the model are closer to the actual sales scenario, thereby providing effective support and guidance for the implementation of sales strategies.

[0144] Example 4 In order to solve the problem that the existing general generative model usually pays too much attention to the content fluency and content format of the pre-sales suggestions when generating pre-sales suggestions, resulting in the lack of pertinence of the generated pre-sales suggestions and the inability to fit the actual sales scenario, thus making it difficult to provide effective support for the implementation of the sales strategy, the embodiment of the present application provides a suggestion generation model training device, the specific structural diagram of the device is as follows Figure 4 As shown, it includes a sample acquisition module 41, a sample processing module 42, an information determination module 43 and an iterative training module 44. The functions of each module are as follows: A sample acquisition module 41, used to acquire a dialogue content sample set and an actual pre-sales suggestion sample set of the dialogue content sample set; A sample processing module 42, used to input the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; An information determination module 43, used to determine the usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and the coverage difference between the key points of the generated pre-sales suggestion sample set and the conversation content sample set; The iterative training module 44 is used to iteratively update the suggestion generation model based on the availability difference and the coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and then stop the iterative update training to obtain the trained suggestion generation model.

[0145] When the device provided in the embodiment of the present application is used, the suggestion generation model can be trained based on the usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and the coverage difference between the key points of the generated pre-sales suggestion sample set and the conversation content sample set. In this way, it can be ensured that the trained suggestion generation model can focus on the key point coverage between the generated pre-sales suggestion and the target conversation information, reduce the situation where the key points in the actual sales scenario are ignored, and then the pre-sales suggestions generated by the model can better cover the key points in the conversation, thereby ensuring that the generated pre-sales suggestions are more targeted and can provide effective support for the implementation of the sales strategy.

[0146] On the other hand, it can also ensure that the trained suggestion generation model can focus on the availability loss between the generated pre-sales suggestions and the actual pre-sales suggestions. In this way, it can ensure that the pre-sales suggestions generated by the model are closer to the actual sales scenario, thereby providing effective support and guidance for the implementation of sales strategies.

[0147] Example 5 Figure 5 A hardware structure diagram of an electronic device for implementing various embodiments of the present application is provided, and the electronic device may include a processor 501 and a memory 502 storing computer program instructions. Specifically, the processor 501 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0148] The memory 502 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 502 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 502 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 502 may be inside or outside the electronic device. In a particular embodiment, the memory 502 may be a non-volatile solid-state memory.

[0149] In one embodiment, the memory 502 may be a read-only memory (ROM). In one embodiment, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or a flash memory, or a combination of two or more of these.

[0150] The processor 501 reads and executes computer program instructions stored in the memory 502 to implement any one of the pre-sales suggestion generating methods or suggestion generating model training methods in the above embodiments.

[0151] In one example, the electronic device may further include a communication interface 505 and a bus 510. Figure 5 As shown, the processor 501, the memory 502, and the communication interface 505 are connected via a bus 510 and communicate with each other.

[0152] The communication interface 505 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0153] Bus 510 includes hardware, software or both, coupling the components of the electronic device to each other. For example, but not limitation, the bus may include an accelerated graphics port (AGP) or other graphics bus, an enhanced industrial standard architecture (EISA) bus, a front-side bus (FSB), a hypertransport (HT) interconnect, an industrial standard architecture (ISA) bus, an infinite bandwidth interconnect, a low pin count (LPC) bus, a memory bus, a microchannel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standard association local (VLB) bus or other suitable bus or a combination of two or more of these. Where appropriate, bus 510 may include one or more buses. Although the present application embodiment describes and illustrates a specific bus, the present application considers any suitable bus or interconnect.

[0154] In addition, in combination with the pre-sale suggestion generation method or suggestion generation model training method in the above embodiments, the present application embodiment can provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any of the pre-sale suggestion generation methods or suggestion generation model training methods in the above embodiments is implemented.

[0155] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0156] The above is only a specific implementation example of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described system, module and unit can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application.

Claims

1. A method for generating pre-sales suggestions, characterized in that: include: Obtain the conversation content between the agent and the customer related to the business to be marketed; Inputting the conversation content into a suggestion generation model to obtain a pre-sales suggestion corresponding to the business to be marketed; Wherein, when training the suggestion generation model, minimizing the sum of the availability loss value and the key point coverage loss value is used as the training goal; The usability loss value is used to characterize the usability difference between the actual pre-sales suggestion sample set and the generated pre-sales suggestion sample set generated by the suggestion generation model based on the conversation content sample set; The key point coverage loss value is used to characterize the difference between the coverage of the expected key points of the pre-sales suggestions by the generative pre-sales suggestion sample set and the conversation content sample set.

2. The method according to claim 1, characterized in that The availability loss value is determined in the following manner: Calculating a first usability score of the generative pre-sales suggestion sample set by using a preset usability function according to the conversation content sample set and the generative pre-sales suggestion sample set; Calculating a second availability score of the actual pre-sale suggestion sample set by using the preset availability function according to the conversation content sample set and the actual pre-sale suggestion sample set; The availability loss value is determined based on an availability difference between the first availability score and the second availability score.

3. The method according to claim 1, characterized in that The key point coverage loss value is determined in the following manner: Determining a set of key points of the conversation content sample set; The key point coverage loss value is determined according to the coverage difference between the keywords of the generative pre-sales suggestion sample set and the key point set.

4. The method according to claim 1, characterized in that Before inputting the conversation content into a suggestion generation model to obtain a pre-sale suggestion corresponding to the business to be marketed, the method further includes: Acquire the dialogue content sample set and the actual pre-sales suggestion sample set of the dialogue content sample set; Inputting the conversation content sample set into the suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; Determining a usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and a coverage difference between key points of the generated pre-sales suggestion sample set and the conversation content sample set; The suggestion generation model is iteratively updated and trained based on the availability difference and the coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and the iterative update training is stopped to obtain the trained suggestion generation model.

5. The method according to claim 1, characterized in that Get the conversation content between the agent and the customer related to the business to be marketed, including: Acquire the conversation between the agent and the customer related to the business to be marketed; Inputting the conversation into a conversation content extraction model to obtain the conversation content; The dialogue content extraction model is trained by using a mixed data set consisting of an information extraction instruction data set and a knowledge instruction data set of dialogue samples as model input, and dialogue content samples of the dialogue samples as model output.

6. The method according to claim 5, characterized in that The mixed data set is constructed in the following way: Determining a correlation threshold, information entropy, and a data ratio of an information extraction instruction data set and a knowledge instruction data set in the mixed data set; According to the correlation threshold, the information entropy and the data ratio, data are selected from the information extraction instruction data set and the knowledge instruction data set to form the mixed data set.

7. The method according to any one of claims 1 to 6, characterized in that: The conversation content includes at least one of the following: The customer's objection to the business to be marketed; The reason why the customer did not renew the business to be marketed; The quotation comparison information of the business to be marketed.

8. A method for training a suggestion generation model, characterized in that: include: Acquire the dialogue content sample set and the actual pre-sales suggestion sample set of the dialogue content sample set; Inputting the conversation content sample set into a suggestion generation model to obtain a generative pre-sales suggestion sample set of the conversation content sample set; Determining a usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and a coverage difference between key points of the generated pre-sales suggestion sample set and the conversation content sample set; The suggestion generation model is iteratively updated and trained based on the availability difference and the coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and the iterative update training is stopped to obtain the trained suggestion generation model.

9. A pre-sale suggestion generating device, characterized in that: It includes an information acquisition module and a suggestion generation module, where: An information acquisition module is used to acquire the conversation content between the agent and the customer related to the business to be marketed; A suggestion generation module, used for inputting the conversation content into a suggestion generation model to obtain a pre-sales suggestion corresponding to the business to be marketed; Wherein, when training the suggestion generation model, minimizing the sum of the availability loss value and the key point coverage loss value is used as the training goal; The usability loss value is used to characterize the usability difference between the actual pre-sales suggestion sample set and the generated pre-sales suggestion sample set generated by the suggestion generation model based on the conversation content sample set; The key point coverage loss value is used to characterize the difference between the coverage of the expected key points of the pre-sales suggestions by the generative pre-sales suggestion sample set and the conversation content sample set.

10. A suggestion generation model training device, characterized in that: It includes a sample acquisition module, a sample processing module, an information determination module and an iterative training module, wherein: A sample acquisition module, used for acquiring the conversation content sample set and the actual pre-sale suggestion sample set of the conversation content sample set; A sample processing module, used for inputting the dialogue content sample set into a suggestion generation model to obtain a generative pre-sales suggestion sample set of the dialogue content sample set; An information determination module, configured to determine a usability difference between the generated pre-sales suggestion sample set and the actual pre-sales suggestion sample set, and a coverage difference between key points of the generated pre-sales suggestion sample set and the conversation content sample set; An iterative training module is used to iteratively update the suggestion generation model based on the availability difference and the coverage difference until the latest coverage difference and the latest availability difference of the suggestion generation model after the iterative update training meet the preset minimization requirements, and stop the iterative update training to obtain the trained suggestion generation model.

11. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the pre-sales suggestion generating method according to any one of claims 1 to 7 or the suggestion generating model training method according to claim 8 are implemented.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the pre-sales advice generation method according to any one of claims 1 to 7 or the advice generation model training method according to claim 8 are implemented.

13. A computer program product, characterized in that The method comprises a computer program, which, when executed by a processor, implements the pre-sale suggestion generating method according to any one of claims 1 to 7, or the steps of the suggestion generating model training method according to claim 8.

Images