Dialogue generation method and device, equipment and medium

Abstract

The embodiment of the invention provides a dialogue generation method and device, equipment and a medium, and relates to the technical field of artificial intelligence. The method comprises the steps of obtaining a current question text; inputting the current question text into a large language model of a routing agent, and determining an intention category corresponding to the current question text through the large language model of the routing agent; and inputting the current question text into the large language model of the intelligent agent corresponding to the intention category, and generating a current answer text corresponding to the current question text, the historical question text and the historical answer text through the large language model of the intelligent agent corresponding to the intention category. Therefore, the dialogue content accurately matched with the customer can be continuously and stably generated through the big language model of the intelligent agent corresponding to the intention category, so that the sales conversion rate of insurance products can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a dialogue generation method, apparatus, device, and medium. Background Technology

[0002] In the insurance industry, the dialogue between agents and customers is a core element determining sales conversion rates. Currently, the generation of these dialogues heavily relies on the agent's individual understanding of insurance products, sales experience, and adaptability.

[0003] However, as insurance products become increasingly complex and customer needs become more diverse, it is difficult for agents to continuously and reliably generate conversations that are precisely matched with customers. This directly restricts the further improvement of the sales conversion rate of insurance products. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a dialogue generation method, apparatus, device, and medium that can improve the sales conversion rate of insurance products.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] In a first aspect, this application discloses a dialogue generation method, the method comprising:

[0007] Get the current problem text;

[0008] The current question text is input into the large language model of the routing agent, and the intent category corresponding to the current question text is determined by the large language model of the routing agent.

[0009] The current question text is input into the large language model of the agent corresponding to the intent category, and the current answer text corresponding to the current question text, historical question text, and historical answer text is generated through the large language model of the agent corresponding to the intent category.

[0010] Optionally, inputting the current problem text into the large language model of the routing agent includes:

[0011] The current problem text is standardized and rewritten using a problem rewriting model to obtain the current standard problem;

[0012] Determine the semantic similarity between the current standard problem and each preset standard problem in the preset standard problem set;

[0013] If the semantic similarity is less than or equal to the preset similarity threshold, then the current problem text is input into the large language model of the routing agent.

[0014] Optionally, the step of inputting the current question text into the large language model of the agent corresponding to the intent category, and generating the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category, includes:

[0015] The current question text is input into the large language model of the agent corresponding to the demand mining category. The large language model of the agent corresponding to the demand mining category determines whether the current question text, historical question text, and historical answer text include customer underwriting information. If yes, a risk assessment is performed on the customer underwriting information. After obtaining the risk assessment result, a first response script including the risk assessment result is generated as the current answer text. If no, a second response script is generated to guide the customer to provide customer underwriting information as the current answer text.

[0016] Optionally, the step of inputting the current question text into the large language model of the agent corresponding to the intent category, and generating the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category, includes:

[0017] The current question text is input into the large language model of the agent corresponding to the product recommendation category. Through the large language model of the agent corresponding to the product recommendation category, at least one target insurance product that matches the current question text and the user's historical profile is selected from the preset insurance product set, and the current answer text including the target insurance product is generated. The user's historical profile is constructed based on historical question text and historical answer text.

[0018] Optionally, the step of inputting the current question text into the large language model of the agent corresponding to the intent category, and generating the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category, includes:

[0019] The current question text is input into the large language model of the agent corresponding to the product service explanation category. The current question text is decomposed into N sub-questions using the large language model of the agent corresponding to the product service explanation category; N is a positive integer. The corresponding business tool is called to execute the N sub-questions to obtain N original knowledge fragments. The N original knowledge fragments are concatenated to obtain a concatenated knowledge fragment. Based on the concatenated knowledge fragment, historical question texts, and historical answer texts, the corresponding current answer text is generated.

[0020] Secondly, this application discloses a dialogue generation device, the device comprising: a text acquisition module, an intent recognition module, and a text generation module;

[0021] The text acquisition module is used to acquire the current question text;

[0022] The intent recognition module is used to input the current question text into the large language model of the routing agent, and determine the intent category corresponding to the current question text through the large language model of the routing agent;

[0023] The text generation module is used to input the current question text into the large language model of the agent corresponding to the intent category, and generate the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category.

[0024] Optionally, the intent recognition module is specifically used to: standardize and rewrite the current question text using a question rewriting model to obtain a current standard question; determine the semantic similarity between the current standard question and each preset standard question in a preset set of standard questions; if the semantic similarity is less than or equal to a preset similarity threshold, then input the current question text into the large language model of the routing agent.

[0025] Optionally, the text generation module is specifically used to: input the current question text into the large language model of the agent corresponding to the demand mining category; and determine, through the large language model of the agent corresponding to the demand mining category, whether the current question text, historical question text, and historical answer text include customer underwriting information items; if so, after performing a risk assessment on the customer underwriting information items and obtaining the risk assessment result, generate a first response script including the risk assessment result as the current answer text; if not, generate a second response script to guide the customer to provide customer underwriting information items as the current answer text.

[0026] Optionally, the text generation module is specifically used to: input the current question text into the large language model of the agent corresponding to the product recommendation category; through the large language model of the agent corresponding to the product recommendation category, select at least one target insurance product from the preset insurance product set that matches the current question text and the user's historical profile, and generate the current answer text including the target insurance product; the user's historical profile is constructed based on historical question texts and historical answer texts.

[0027] Optionally, the text generation module is specifically used to: input the current question text into the large language model of the agent corresponding to the product service explanation category; decompose the current question text into N sub-questions using the large language model of the agent corresponding to the product service explanation category; N is a positive integer; call the corresponding business tool to execute the N sub-questions to obtain N original knowledge fragments; concatenate the N original knowledge fragments to obtain a concatenated knowledge fragment; and generate the corresponding current answer text based on the concatenated knowledge fragment, historical question text, and historical answer text.

[0028] Thirdly, this application discloses a dialogue generation device, the device comprising: a memory and a processor;

[0029] The memory is used to store programs;

[0030] The processor is configured to execute the program to implement the various steps of the dialogue generation method as described in the first aspect.

[0031] Fourthly, this application discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the various steps of the dialogue generation method as described in the first aspect.

[0032] Compared with the prior art, this application has the following beneficial effects:

[0033] This application provides a dialogue generation method, apparatus, device, and medium. The method includes: acquiring current question text; inputting the current question text into a large language model of a routing agent, determining the intent category corresponding to the current question text through the large language model of the routing agent; inputting the current question text into a large language model of an agent corresponding to the intent category, generating current answer text corresponding to the current question text, historical question texts, and historical answer texts through the large language model of the agent corresponding to the intent category. Therefore, by using the large language model of the agent corresponding to the intent category, dialogue content accurately matched to customers can be continuously and stably generated, thereby improving the sales conversion rate of insurance products. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart illustrating a dialogue generation method provided in an embodiment of this application;

[0036] Figure 2 A schematic diagram illustrating a dialogue generation method provided in an embodiment of this application;

[0037] Figure 3 A schematic diagram of a dialogue generation device provided in an embodiment of this application;

[0038] Figure 4 This is a schematic diagram of a computer-readable medium provided in an embodiment of this application. Detailed Implementation

[0039] As described earlier, current dialogue generation heavily relies on agents' understanding of insurance products, sales experience, and on-the-spot adaptability. However, with insurance products becoming increasingly complex and customer needs becoming more diverse, relying solely on agents to manually generate dialogue content that accurately matches customers is no longer sufficient. This directly hinders further improvements in insurance product sales conversion rates.

[0040] The inventors, through research, proposed a dialogue generation method, apparatus, device, and medium. The method includes: acquiring the current question text; inputting the current question text into a large language model of a routing agent, determining the intent category corresponding to the current question text through the large language model of the routing agent; inputting the current question text into the large language model of the agent corresponding to the intent category, generating the current answer text corresponding to the current question text, historical question texts, and historical answer texts through the large language model of the agent corresponding to the intent category. Therefore, by using the large language model of the agent corresponding to the intent category, dialogue content accurately matched to customers can be continuously and stably generated, thereby improving the sales conversion rate of insurance products.

[0041] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0042] See Figure 1 The figure is a flowchart of a dialogue generation method provided in an embodiment of this application. The method includes:

[0043] S101: Get the current problem text.

[0044] The current issue text refers to the natural language expression unit that is most recently entered by the customer and requires a response.

[0045] It is understood that the current question text in this application is not limited to grammatically interrogative sentences, but refers to any respondable customer utterance. For example, the current question text could be "What diseases does this critical illness insurance cover?", "I think the premium is a bit high.", "Could you explain in more detail? I didn't understand what you meant.", etc.

[0046] S102: Input the current question text into the large language model of the routing agent, and determine the intent category corresponding to the current question text through the large language model of the routing agent.

[0047] First, addressing the non-standard nature of the current question text (e.g., colloquialism, incompleteness, ambiguity, or dialectal features), a question rewriting model is used to standardize and rewrite the current question text, resulting in a standard question. For example, "How to bind a bank card?" is rewritten as "How to bind a personal bank card account?", and "Is the disease I mentioned last time covered?" is rewritten as "Is the disease I inquired about before covered?".

[0048] Subsequently, the semantic similarity between the current standard question and each preset standard question in the preset standard question set is determined. For example, the cosine similarity or Euclidean distance between the vector of the current standard question and the vectors of each preset standard question in the preset standard question set can be determined. It is understood that the preset standard question set covers more than 80% of common consultation points in insurance sales (such as rates, coverage, application process, claims materials, etc.).

[0049] Next, it is determined whether the semantic similarity is less than or equal to a preset similarity threshold (e.g., 0.9). If not, a preset standard question corresponding to the semantic similarity above the preset similarity threshold is identified, and a standard answer corresponding to that preset standard question is determined from the set of standard dialogues. Thus, for high-frequency, common, and standard current question texts, a millisecond-level ultra-fast response can be achieved, avoiding unnecessary computational costs from calling large language models.

[0050] If so, the current problem text is input into the large language model of the routing agent (e.g., qwen2.5-72B model).

[0051] The large language model of the routing agent takes the current question text and the first prompt word as input, and outputs the intent category corresponding to the current question text. For example, the first prompt word could be "You are a professional insurance sales dialogue analyst. Please help me determine the corresponding intent category from the preset intent category set based on the current question text. The preset intent category set includes: {understanding customers category, demand mining category, product recommendation category, product service explanation category, order facilitation category, and human intervention category}".

[0052] It should be noted that if the current issue text is too vague, exceeds the preset scope, or contains strong emotions (such as angry complaints), and the intent category corresponding to the current issue text cannot be determined in the preset intent category set, then the intent category corresponding to the current issue text will be forcibly determined to be the manual intervention category.

[0053] S103: Input the current question text into the large language model of the agent corresponding to the intent category, and generate the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category.

[0054] See Figure 2 This figure is a schematic diagram of a dialogue generation method provided in an embodiment of this application. Figure 2 As can be seen, the intent categories include customer understanding, demand mining, product recommendation, product service explanation, order facilitation, and human intervention. Therefore, the corresponding intelligent agents for these intent categories are: customer understanding agent, demand mining agent, product recommendation agent, product service explanation agent, order facilitation agent, and human intervention agent.

[0055] Scenario 1:

[0056] If the intent category corresponding to the current question text is "Understanding the Customer," then the current question text is input into the large language model of the "Understanding the Customer" agent. The core task of the large language model of the "Understanding the Customer" agent is to guide the dialogue using the current text question as a starting point to obtain the background information necessary for building an accurate user profile.

[0057] Therefore, understanding the large language model of the client's intelligent agent involves the following steps:

[0058] First, user history profiles are obtained. These profiles are constructed using information extraction and intent understanding technologies based on historical question and answer texts.

[0059] Subsequently, based on the current question text and the user's historical profile, the current answer text with a corresponding wording style to the user's historical profile is generated.

[0060] See Table 1, which is a schematic table of the first type of current answer text provided in the embodiments of this application.

[0061] Table 1

[0062]

[0063] Scenario 2:

[0064] If the intent category corresponding to the current question text is a demand mining category, then the current question text is input into the large language model of the demand mining agent. The core task of the large language model of the demand mining agent is to transform the customer's vague demands into clear, quantifiable, and insurable risk protection demands.

[0065] Therefore, the large language model of the demand mining agent will perform the following steps:

[0066] First, determine whether the current question text, historical question texts, and historical answer texts include customer underwriting information. Customer underwriting information is core data necessary for insurance risk assessment and product matching, and includes at least one of the following: age, health status, and intended product.

[0067] If so, then: First, invoke the underwriting tool to perform a risk assessment on the customer's underwriting information items and obtain the risk assessment results. Then, generate a first response script including the risk assessment results as the current answer text.

[0068] See Table 2, which is a schematic table of the second type of current answer text provided in the embodiments of this application.

[0069] Table 2

[0070]

[0071] If not, then: generate a second response script to guide the customer to provide customer underwriting information as the current answer text.

[0072] See Table 3, which is an illustration of the third type of current answer text provided in the embodiments of this application.

[0073] Table 3

[0074]

[0075] Scenario 3:

[0076] If the intent category corresponding to the current question text is product recommendation, then the current question text is input into the large language model of the product recommendation agent. The core task of the large language model of the product recommendation agent is to present one or more specific and feasible solutions to the customer based on the user history profiles accumulated in the early stage.

[0077] Therefore, the large language model of the product recommendation agent will perform the following steps:

[0078] First, user history profiles are obtained. These profiles are constructed using information extraction and intent understanding technologies based on historical question and answer texts.

[0079] Subsequently, from a pre-defined set of insurance products, at least one target insurance product is selected that matches the current question text and the user's historical profile. This pre-defined set of insurance products is a structured knowledge base of insurance products, including not only the names of the products but also rich structured tags such as suitable age range, recommended occupational types, and benefits for those with high blood pressure.

[0080] Finally, generate the current answer text, which includes the target insurance product.

[0081] See Table 4, which is an illustration of the fourth type of current answer text provided in the embodiments of this application.

[0082] Table 4

[0083]

[0084] Scenario 4:

[0085] If the intent category corresponding to the current question text is the product service explanation category, then the current question text is input into the large language model of the product service explanation agent. The core task of the large language model of the product service explanation agent is to respond accurately, reliably, clearly, and in a friendly manner to professional questions involving specific data and rules.

[0086] Then, the large language model of the product service explanation agent will perform the following steps:

[0087] First, the current problem text is broken down into N sub-problems (N is a positive integer). Each sub-problem corresponds to a business tool, which includes a product knowledge Q&A tool, a premium calculation tool, a product link tool, and an underwriting tool. The product knowledge Q&A tool is used to query static factual information such as waiting periods, coverage, exclusions, and claims procedures. The premium calculation tool outputs the accurate premium amount. The product link tool outputs the latest and valid links. The underwriting tool is used to pre-assess specific health disclosures.

[0088] Secondly, the corresponding business tools are invoked to execute N sub-problems, resulting in N raw knowledge fragments. Understandably, these raw knowledge fragments are typically structured data (e.g., JSON) or text in a specific format.

[0089] Subsequently, the N original knowledge fragments are spliced ​​together to obtain the spliced ​​knowledge fragment.

[0090] Finally, the current answer text is determined by piecing together knowledge fragments, historical question texts, and historical answer texts.

[0091] It is understandable that after obtaining the pieced-together knowledge fragments, these fragments can be summarized, deduplicated, paraphrased in a conversational style, and have key points highlighted to obtain the processed knowledge fragments. Furthermore, the corresponding current answer text can be generated based on these processed knowledge fragments, historical question texts, and historical answer texts. This application does not impose any limitations on this.

[0092] See Table 5, which is an illustration of the fifth type of current answer text provided in the embodiments of this application.

[0093] Table 5

[0094]

[0095] Scenario 5:

[0096] If the intent category corresponding to the current question text is order closing, then the current question text is input into the large language model of the order closing agent. The core task of the large language model of the order closing agent is to diagnose the customer's real concerns, formulate targeted persuasion strategies, and ultimately generate communication scripts that can effectively help customers overcome decision-making thresholds.

[0097] Therefore, the large language model that facilitates the order will perform the following steps:

[0098] First, based on the current question text, historical question texts, and historical answer texts, information about the thought process is generated.

[0099] Secondly, the thought process information is broken down into N sub-problems (N is a positive integer), and each sub-problem corresponds to a business tool, including product knowledge Q&A tool, premium calculation tool, product linking tool, and underwriting tool.

[0100] Subsequently, the N original knowledge fragments are spliced ​​together to obtain the spliced ​​knowledge fragment.

[0101] Finally, the corresponding current answer text is generated based on the spliced ​​knowledge fragments, historical question texts, and historical answer texts.

[0102] It is understandable that after obtaining the pieced-together knowledge fragments, these fragments can be summarized, deduplicated, paraphrased in a conversational style, and have key points highlighted to obtain processed knowledge fragments. Furthermore, the current answer text can be derived based on these processed knowledge fragments, historical question texts, and historical answer texts. This application does not impose any limitations on this aspect.

[0103] For example, if the historical question text and historical answer text are: "Customer: Please help me check the waiting period for the product 'Family Guardian Critical Illness Insurance' and how much the premium is? Agent: Based on your situation (32 years old), if you purchase 'Family Guardian Critical Illness Insurance', the waiting period is 90 days and the premium is 9450 yuan per year."; and the current question text is: "I feel the price pressure is still a bit too much, I'll think about it some more." Then, the generated thought process information can be: "The customer expressed price pressure and decision-making delay. Persuasion strategies should include: 1) Cost breakdown (What is the average daily cost?); 2) Providing alternatives (Does monthly payment support it?). Then, the current answer text in Table 6 below can be generated. See Table 6, which is an illustration of the sixth type of current answer text provided in this application embodiment.

[0104] Table 6

[0105]

[0106] Situation 6:

[0107] If the intent category corresponding to the current question text is the human intervention category, then the current question text is input into the large language model of the human intervention agent. The large language model of the human intervention agent will then perform the following steps:

[0108] Based on the current question text, historical question texts, and historical answer texts, the prompt word template is filled in to generate the current answer text. See Table 7, which is an illustration of the seventh type of current answer text provided in this application embodiment.

[0109] Table 7

[0110]

[0111] In summary, the embodiments of this application provide a dialogue generation method. This method can continuously and stably generate dialogue content that accurately matches customers through a large language model of an agent corresponding to an intent category, thereby improving the sales conversion rate of insurance products.

[0112] See Figure 3 The figure is a schematic diagram of a dialogue generation device provided in an embodiment of this application. The dialogue generation device 300 includes: a text acquisition module 301, an intent recognition module 302, and a text generation module 303;

[0113] Text acquisition module 301 is used to acquire the text of the current question;

[0114] The intent recognition module 302 is used to input the current question text into the large language model of the routing agent, and determine the intent category corresponding to the current question text through the large language model of the routing agent.

[0115] The text generation module 303 is used to input the current question text into the large language model of the agent corresponding to the intent category, and generate the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category.

[0116] In one specific implementation, the intent recognition module 302 is specifically used to: standardize and rewrite the current question text using a question rewriting model to obtain the current standard question; determine the semantic similarity between the current standard question and each preset standard question in the preset standard question set; if the semantic similarity is less than or equal to the preset similarity threshold, then input the current question text into the large language model of the routing agent.

[0117] In one specific implementation, the text generation module 303 is specifically used to: input the current question text into the large language model of the agent corresponding to the demand mining category; and through the large language model of the agent corresponding to the demand mining category, determine whether the current question text, historical question text, and historical answer text include customer underwriting information items; if so, perform a risk assessment on the customer underwriting information items, and after obtaining the risk assessment result, generate a first response script including the risk assessment result as the current answer text; if not, generate a second response script to guide the customer to provide customer underwriting information items as the current answer text.

[0118] In one specific implementation, the text generation module 303 is specifically used to: input the current question text into the large language model of the agent corresponding to the product recommendation category; through the large language model of the agent corresponding to the product recommendation category, select at least one target insurance product from the preset insurance product set that matches the current question text and the user's historical profile, and generate the current answer text including the target insurance product; the user's historical profile is constructed based on historical question texts and historical answer texts.

[0119] In one specific implementation, the text generation module 303 is specifically used to: input the current question text into the large language model of the agent corresponding to the product service explanation category; decompose the current question text into N sub-questions using the large language model of the agent corresponding to the product service explanation category; N is a positive integer; call the corresponding business tool to execute the N sub-questions to obtain N original knowledge fragments; concatenate the N original knowledge fragments to obtain a concatenated knowledge fragment; and generate the corresponding current answer text based on the concatenated knowledge fragment, historical question text, and historical answer text.

[0120] In summary, the embodiments of this application provide a dialogue generation device that can continuously and stably generate dialogue content that accurately matches customers through a large language model of an agent corresponding to an intent category, thereby improving the sales conversion rate of insurance products.

[0121] This application also provides a corresponding dialogue generation device and a computer-readable medium for implementing the dialogue generation method provided in this application.

[0122] The dialogue generation device includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code to cause the device to perform a dialogue generation method according to any embodiment of this application.

[0123] See Figure 4 This figure is a schematic diagram of a computer-readable medium provided in an embodiment of this application. The computer-readable medium 400 stores a computer program 411, which, when executed by a processor, implements the above-described... Figure 1 The steps of the dialogue generation method.

[0124] It should be noted that, in the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0125] It should be noted that the machine-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0126] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0127] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

[0128] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0129] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A dialogue generation method, characterized in that, The method includes: Get the current problem text; The current question text is input into the large language model of the routing agent, and the intent category corresponding to the current question text is determined by the large language model of the routing agent. The current question text is input into the large language model of the agent corresponding to the intent category, and the current answer text corresponding to the current question text, historical question text, and historical answer text is generated through the large language model of the agent corresponding to the intent category.

2. The method according to claim 1, characterized in that, The process of inputting the current problem text into the large language model of the routing agent includes: The current problem text is standardized and rewritten using a problem rewriting model to obtain the current standard problem; Determine the semantic similarity between the current standard problem and each preset standard problem in the preset standard problem set; If the semantic similarity is less than or equal to the preset similarity threshold, then the current problem text is input into the large language model of the routing agent.

3. The method according to claim 1, characterized in that, The step of inputting the current question text into the large language model of the agent corresponding to the intent category, and generating the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category, includes: The current question text is input into the large language model of the agent corresponding to the demand mining category. The large language model of the agent corresponding to the demand mining category determines whether the current question text, historical question text, and historical answer text include customer underwriting information. If yes, a risk assessment is performed on the customer underwriting information. After obtaining the risk assessment result, a first response script including the risk assessment result is generated as the current answer text. If no, a second response script is generated to guide the customer to provide customer underwriting information as the current answer text.

4. The method according to claim 1, characterized in that, The step of inputting the current question text into the large language model of the agent corresponding to the intent category, and generating the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category, includes: The current question text is input into the large language model of the agent corresponding to the product recommendation category. Through the large language model of the agent corresponding to the product recommendation category, at least one target insurance product that matches the current question text and the user's historical profile is selected from the preset insurance product set, and the current answer text including the target insurance product is generated. The user's historical profile is constructed based on historical question text and historical answer text.

5. The method according to claim 1, characterized in that, The step of inputting the current question text into the large language model of the agent corresponding to the intent category, and generating the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category, includes: The current question text is input into the large language model of the agent corresponding to the product service explanation category. The current question text is decomposed into N sub-questions using the large language model of the agent corresponding to the product service explanation category; N is a positive integer. The corresponding business tool is called to execute the N sub-questions to obtain N original knowledge fragments. The N original knowledge fragments are concatenated to obtain a concatenated knowledge fragment. Based on the concatenated knowledge fragment, historical question texts, and historical answer texts, the corresponding current answer text is generated.

6. A dialogue generation device, characterized in that, The device includes: a text acquisition module, an intent recognition module, and a text generation module; The text acquisition module is used to acquire the current question text; The intent recognition module is used to input the current question text into the large language model of the routing agent, and determine the intent category corresponding to the current question text through the large language model of the routing agent; The text generation module is used to input the current question text into the large language model of the agent corresponding to the intent category, and generate the current answer text corresponding to the current question text, historical question text, and historical answer text through the large language model of the agent corresponding to the intent category.

7. The apparatus according to claim 6, characterized in that, The intent recognition module is specifically used to: standardize and rewrite the current question text using a question rewriting model to obtain a current standard question; and determine the semantic similarity between the current standard question and each preset standard question in a preset set of standard questions. If the semantic similarity is less than or equal to the preset similarity threshold, then the current problem text is input into the large language model of the routing agent.

8. The apparatus according to claim 6, characterized in that, The text generation module is specifically used to: input the current question text into the large language model of the agent corresponding to the demand mining category, and determine whether the current question text, historical question text, and historical answer text include customer underwriting information items through the large language model of the agent corresponding to the demand mining category; if so, after performing a risk assessment on the customer underwriting information items and obtaining the risk assessment result, generate a first response script including the risk assessment result as the current answer text; If not, a second response script will be generated to guide the customer to provide customer underwriting information as the current answer text.

9. A dialogue generation device, characterized in that, The device includes: a memory and a processor; The memory is used to store programs; The processor is configured to execute the program to implement the various steps of the dialogue generation method as described in any one of claims 1 to 5.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the various steps of the dialogue generation method as described in any one of claims 1 to 5.

Images