A voice question and answer construction method and device based on a large language model and a medium

Abstract

The application discloses a voice question and answer construction method based on a large language model, comprising the following steps: acquiring enterprise information; constructing an enterprise prompt according to the enterprise information, inputting the enterprise prompt into a large language model to acquire a plurality of first industry knowledge points; performing knowledge condensation and knowledge deduplication on the first industry knowledge points to acquire second industry knowledge points; constructing a questionnaire prompt according to the second industry knowledge points, inputting the questionnaire prompt into the large language model to acquire a dialogue questionnaire, and acquiring enterprise knowledge through the dialogue questionnaire; constructing a dialogue prompt according to the enterprise knowledge, inputting the dialogue prompt into the large language model to acquire a response dialogue; acquiring voice reply information of the response dialogue, and constructing a voice question and answer dialogue library, wherein the voice question and answer dialogue library comprises questions and corresponding voice reply information. The application automatically mines industry knowledge and generates a dialogue questionnaire through a large language model to quickly collect enterprise knowledge, so as to quickly support dialogue cold start.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a voice question and answer construction method and device based on a large language model and a medium. BACKGROUND

[0002] An intelligent voice question and answer system is a system that automatically replies according to user questions. At present, the voice question and answer system is widely applied to enterprise customer service, sales and other scenarios. Since the purchaser usually needs to obtain the product or service information of the enterprise through telephone consultation, the answer quality of the voice question and answer system of the enterprise will directly affect the transaction willingness of the purchaser.

[0003] When the existing enterprise voice question and answer system is constructed, the staff first needs to design a questionnaire for the enterprise. The questionnaire contains product-related questions that the purchaser may mention in the conversation. The enterprise answers the questions by recording to realize the cold start of the voice question and answer system suitable for the enterprise. However, the existing technology often faces the following problems:

[0004] 1. The staff does not have a deep understanding of the enterprise and its industry. The staff needs to spend effort to understand the industry knowledge so that the designed questionnaire is difficult to fit the knowledge points concerned by the actual purchaser.

[0005] 2. The process of constructing the question and answer is complicated, and the enterprise and the staff need to cooperate for many times. The enterprise personnel need to answer the questionnaire and record, which reduces the generation efficiency.

[0006] 3. The questions provided by the staff cannot cover the relatively cold knowledge points, and the needed information cannot be obtained when the purchaser inquires. SUMMARY

[0007] In order to solve the above problems, the first aspect of the present application discloses a voice question and answer construction method based on a large language model, which comprises:

[0008] Obtaining enterprise information through voice interaction or text interaction, wherein the enterprise information comprises an enterprise name, a question name and product information;

[0009] Constructing an enterprise prompt according to the enterprise information, inputting the enterprise prompt into a large language model to obtain a plurality of first industry knowledge points;

[0010] Carrying out knowledge condensation and knowledge deduplication on the first industry knowledge points to obtain second industry knowledge points;

[0011] Constructing a questionnaire prompt according to the second industry knowledge points, inputting the questionnaire prompt into the large language model to obtain a question and answer questionnaire, and obtaining enterprise knowledge through the question and answer questionnaire;

[0012] According to the enterprise knowledge construction dialogue prompt, the dialogue prompt is input into a large language model to obtain a response dialogue;

[0013] Voice reply information of the response dialogue is obtained, and a voice question and answer dialogue library is constructed, the voice question and answer dialogue library including questions and corresponding voice reply information.

[0014] In a further embodiment, the enterprise prompt is used to guide the large language model to inquire about first industry knowledge points related to the product information in an electric sales scenario.

[0015] In a further embodiment, the questionnaire prompt is used to guide the large language model to convert the second industry knowledge points into questions and reference answer elements, generate reference answer samples based on the reference answer elements, and generate a dialogue questionnaire in combination with the questions, reference answer elements, and reference answer samples.

[0016] In a further embodiment, the dialogue prompt is used to guide the large language model to convert the enterprise knowledge into response dialogues based on customer service roles.

[0017] In a further embodiment, when the first industry knowledge points are condensed, a semantic matching model is used to cross-match a plurality of the first industry knowledge points, and a high-frequency first industry knowledge point is selected through a voting mechanism.

[0018] In a further embodiment, when the first industry knowledge points are de-duplicated, the first industry knowledge points after knowledge condensation are repeatedly determined by the semantic matching model to remove identical knowledge points in the first industry knowledge points.

[0019] In a further embodiment, when the voice question and answer dialogue library is running, knowledge discovery is performed at a set period, the knowledge discovery including: collecting questions and clustering them by the semantic matching model when the voice question and answer dialogue library is running, obtaining third industry knowledge points, and supplementing the third industry knowledge points to the first industry knowledge points.

[0020] In a further embodiment, after the first industry knowledge points are condensed, de-duplicated, and second industry knowledge points are obtained, the method further includes: constructing a question prompt according to the second industry knowledge points, the question prompt being used to guide the large language model to obtain a plurality of question forms corresponding to the second industry knowledge points; and supplementing the plurality of question forms of the second industry knowledge points to the voice question and answer dialogue library.

[0021] In a further embodiment, when the voice reply information of the response script is acquired, the TTS interface is used to convert the text form of the response script into the voice form of the voice reply information.

[0022] The second aspect of the present application discloses a voice question and answer construction device based on a large language model, and the device comprises:

[0023] An information acquisition module is configured to acquire enterprise information through voice interaction or text interaction, wherein the enterprise information comprises an enterprise name, a script name, and product information.

[0024] A first knowledge point module is configured to construct an enterprise prompt according to the enterprise information, input the enterprise prompt into a large language model, and acquire a plurality of first industry knowledge points.

[0025] A second knowledge point module is configured to condense and remove duplicate knowledge from the first industry knowledge points to acquire second industry knowledge points.

[0026] A questionnaire generation module is configured to construct a questionnaire prompt according to the second industry knowledge points, input the questionnaire prompt into a large language model, acquire a script questionnaire, and acquire enterprise knowledge through the script questionnaire.

[0027] A script generation module is configured to construct a script prompt according to the enterprise knowledge, input the script prompt into a large language model, and acquire a response script.

[0028] A script construction module is configured to acquire voice reply information of the response script, construct a voice question and answer script library, and the voice question and answer script library comprises questions and corresponding voice reply information.

[0029] The third aspect of the present application discloses a computer storage medium, and the computer storage medium stores computer instructions.

[0030] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0031] In the embodiment of the present application, first, enterprise information is acquired to determine the industry knowledge of the enterprise; based on the large voice model, the enterprise prompt is constructed to obtain the first industry knowledge point, and the first industry knowledge point is generated according to the first industry knowledge point; the advantage that the large language model includes a large amount of industry knowledge is fully utilized, so that the generated first industry knowledge point is more comprehensive and appropriate, and the process of understanding industry knowledge by the staff is avoided, and the production efficiency of the voice question and answer technique is improved. The enterprise knowledge is collected through the technique questionnaire, and the enterprise can supplement the enterprise knowledge to the corresponding question in the form of voice or text, and quickly supports the cold start of the technique. Through knowledge condensation and knowledge deduplication, the knowledge point is more in line with the hot spot concerned by the purchaser, and it is ensured that the collected enterprise knowledge can be used to answer most of the questions. The enterprise knowledge is converted into a formal answer through the technique prompt, and is further converted into voice reply information in the form of voice to answer the questions of the purchaser, effectively reducing the burden of the enterprise, and helping the enterprise to quickly build product techniques. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0033] Figure 1 is a flow diagram of a voice question and answer construction method based on a large language model disclosed by the embodiment of the present application;

[0034] Figure 2 is a structural diagram of a voice question and answer construction device based on a large language model disclosed by the embodiment of the present application;

[0035] Figure 3 is a structural diagram of a computer storage medium disclosed by the embodiment of the present application. DETAILED EMBODIMENT

[0036] In order to make the personnel in the technical field better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0037] The terms "first", "second", and the like in the description and claims of the present application and above drawings are used to distinguish different objects, and are not used to describe a particular order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or end including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or end.

[0038] Reference herein to "embodiments" means that the particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It is explicitly and implicitly understood that the embodiments described herein can be combined with other embodiments.

[0039] The large language model in the embodiments of the present application refers to a deep learning model trained using a large amount of text data, which can generate natural language text or understand the meaning of language text. The large language model can handle a variety of natural language tasks such as text classification, question answering, dialogue, etc., and is an important way to artificial intelligence. Existing large language models such as ChatGPT and Wenxin Yanyan provide interfaces that can be called externally.

[0040] The prompt is an artificial intelligence prompt word, which is a method of using natural language to guide or stimulate artificial intelligence models to complete specific tasks. The role of the prompt is to prompt the context of the input information and the parameter information of the input model for the large language model. When training supervised learning or unsupervised learning models, the prompt can help the model better understand the intent of the input and make appropriate responses.

[0041] The application discloses a voice question and answer construction method and device based on a large language model and medium. In the embodiment of the application, first, enterprise information is acquired to determine the industry knowledge of the enterprise; based on the large voice model, an enterprise prompt is constructed to obtain first industry knowledge points, and a dialogue technique questionnaire is generated according to the first industry knowledge points; the large language model fully utilizes the advantage that it includes massive industry knowledge, so that the generated first industry knowledge points are more comprehensive and appropriate, the process of staff understanding industry knowledge is saved, and the production efficiency of the voice question and answer dialogue technique is improved. The enterprise knowledge is collected through the dialogue technique questionnaire, the enterprise can supplement the enterprise knowledge to the corresponding question in the form of voice or text, and quickly supports the cold start of the dialogue technique. Through knowledge condensation and knowledge deduplication, the knowledge points are more in line with the hotspots concerned by the purchasers, and it is ensured that the collected enterprise knowledge can be used to answer most questions. The enterprise knowledge is converted into a formal answer through the dialogue technique prompt, and is further converted into voice reply information in the form of voice to answer the questions of the purchasers, effectively reducing the burden of the enterprise and helping the enterprise to quickly construct product dialogue techniques. The following will be described in detail.

[0042] Embodiment one

[0043] Please refer to Figure 1 , Figure 1 is a flowchart of a voice question and answer construction method based on a large language model disclosed by the embodiment of the application. As shown in Figure 1 , the voice question and answer construction method based on a large language model can include the following operations:

[0044] S1. Acquire enterprise information through voice interaction or text interaction, and the enterprise information includes enterprise name, dialogue technique name and product information;

[0045] S2. Construct an enterprise prompt according to the enterprise information, and input the enterprise prompt into a large language model to obtain a plurality of first industry knowledge points; it can be understood that the enterprise prompt is used to guide the large language model to inquire about the first industry knowledge points related to product information in the electric sales scene, and in this step, the large language model plays the role of a purchaser through the enterprise prompt, and inquires about the first industry knowledge points in the electric sales scene for the goods or services sold by the enterprise.

[0046] S3. The first industry knowledge point is knowledge condensation and knowledge deduplication to obtain the second industry knowledge point; it can be understood that part of the first industry knowledge points generated by the large language model belongs to the cold industry knowledge, and the purchasing personnel in the electric sales scene usually do not care and consult, and it will not only reduce the enterprise experience, but also reduce the accuracy of the answer to the purchasing personnel. Therefore, it is still necessary to condense the knowledge of the first industry knowledge point. When condensing the knowledge of the first industry knowledge point, the large language model is inquired several times to obtain several first industry knowledge points, the semantic matching model is used to cross-match the several first industry knowledge points, and the high-frequency first industry knowledge points are screened out through the voting mechanism. The definition of high frequency can be set in advance to extract the first industry knowledge point when the number of occurrences or the proportion of the first industry knowledge point reaches a certain threshold. Through knowledge condensation, the questionnaires are more focused on industry hot knowledge, and the prepared questionnaires are more in line with the actual concerns of the purchasing personnel.

[0047] Since the first industry knowledge point contains knowledge points with different expressions but similar or identical meanings, it is still necessary to deduplicate the knowledge of the first industry knowledge point. When deduplicating the knowledge of the first industry knowledge point, the first industry knowledge point after knowledge condensation is repeatedly determined by the semantic matching model to remove the same knowledge points in the first industry knowledge point.

[0048] S4. Construct a questionnaire prompt according to the second industry knowledge point, input the questionnaire prompt into the large language model to obtain the questionnaires, and obtain enterprise knowledge through the questionnaires; specifically, the questionnaire prompt is used to guide the large language model to convert the second industry knowledge point into a question and a reference answer element, generate a reference answer sample based on the reference answer element, and generate a questionnaires by combining the question, the reference answer element and the reference answer sample. In the questionnaire prompt, the knowledge points are converted into corresponding knowledge point questions based on the role of the third party consultant. Further, the enterprise provides enterprise knowledge to the large language model by answering the questionnaires prepared by the large language model, which is used to construct accurate reply questionnaires. Further, by providing reference answer elements and reference answer samples respectively, the enterprise can help the enterprise quickly generate reply questionnaires for different questions.

[0049] S5. Construct a question prompt according to the enterprise knowledge, input the question prompt into the large language model to obtain the reply questionnaires; it can be understood that since the enterprise knowledge obtained is not accurate and clear enough in tone and form, it is difficult to meet the requirements of the questionnaires as an answer to the purchasing personnel, and the large language model still needs to organize the enterprise knowledge to obtain formal reply content. Specifically, the question prompt is used to guide the large language model to convert the enterprise knowledge into a reply questionnaires based on the role of the customer service.

[0050] S6. Obtain voice reply information of the response dialogue, and construct a voice question and answer dialogue library, which includes questions and corresponding voice reply information. When obtaining the voice reply information of the response dialogue, the text form of the response dialogue is converted into voice form voice reply information through a TTS interface. It can be understood that the TTS interface is only one feasible solution in the embodiment, and other technical solutions capable of converting text into voice are also within the protection scope of the present application. The constructed voice question and answer dialogue library can be directly used for the enterprise to answer the questions of the purchaser. The voice reply information is automatically generated, which effectively reduces the workload of the enterprise and facilitates the rapid generation of dialogues suitable for the products or services of the enterprise.

[0051] In an optional embodiment, after the knowledge condensation and knowledge deduplication of the first industry knowledge point and the acquisition of the second industry knowledge point, the method further comprises: constructing a question prompt according to the second industry knowledge point, the question prompt being used to guide the large language model to acquire a plurality of question forms corresponding to the second industry knowledge point; and supplementing the plurality of question forms of the second industry knowledge point to the voice question and answer dialogue library. Specifically, through the above-mentioned solution, the common question methods of the purchasers corresponding to each knowledge point are enriched, and the reliability of the voice question and answer dialogue library is further improved, thereby avoiding the problem that the voice question and answer dialogue library cannot call the voice reply information corresponding to the knowledge point when facing different question methods of the purchasers based on the same knowledge point.

[0052] In an optional embodiment, in order to be able to acquire the knowledge points that may be concerned by the purchasers in actual application, when the voice question and answer dialogue library is running, knowledge discovery is performed at a set period, and the knowledge discovery comprises: collecting questions and clustering through a semantic matching model to acquire third industry knowledge points when the voice question and answer dialogue library is running, and supplementing the third industry knowledge points to the first industry knowledge points. Through automatic analysis and induction of the large language model on the knowledge points concerned by the users in actual application, the response dialogue is iterated, and the effect of the response dialogue in actual application is improved.

[0053] The present application has at least the following beneficial effects:

[0054] (1) The present application automatically mines industry knowledge through a large language model and generates a dialogue questionnaire to quickly collect enterprise knowledge, thereby quickly supporting dialogue cold start.

[0055] (2) Through knowledge condensation, the dialogue questionnaire is more focused on industry hot knowledge, and the produced dialogue is more suitable for the actual concerns of the purchasers.

[0056] (3) The voice reply information is automatically generated, which effectively reduces the workload of the enterprise and facilitates the rapid generation of dialogues suitable for the products or services of the enterprise.

[0057] It can be seen that the present applicationFigure 1 The described large language model-based voice question and answer construction method first acquires enterprise information to determine the industry knowledge of the enterprise; based on the large voice model, the enterprise prompt is constructed to obtain the first industry knowledge point, and the dialogue technique questionnaire is generated according to the first industry knowledge point; the advantages of the large language model including a large amount of industry knowledge are fully utilized, so that the generated first industry knowledge point is more comprehensive and appropriate, the process of understanding industry knowledge by the staff is saved, and the production efficiency of the voice question and answer dialogue technique is improved. The enterprise knowledge is collected through the dialogue technique questionnaire, and the enterprise can supplement the enterprise knowledge to the corresponding question in the form of voice or text to quickly support the cold start of the dialogue technique. Through knowledge condensation and knowledge deduplication, the knowledge points are more in line with the hot spots concerned by the purchasers, and it is ensured that the collected enterprise knowledge can be used to answer most of the questions. The enterprise knowledge is converted into a formal answer through the dialogue technique prompt, and is further converted into voice reply information in the form of voice to answer the questions of the purchasers, effectively reducing the burden of the enterprise and helping the enterprise to quickly construct product dialogue techniques.

[0058] Embodiment two

[0059] Please refer to Figure 2 , Figure 2 is a structural schematic diagram of a large language model-based voice question and answer construction device disclosed by an embodiment of the application. As shown in Figure 2 , the large language model-based voice question and answer construction device can include:

[0060] The information acquisition module 201 is configured to acquire enterprise information through voice interaction or text interaction, and the enterprise information includes enterprise name, dialogue technique name and product information;

[0061] The first knowledge point module 202 is configured to construct an enterprise prompt according to the enterprise information, and input the enterprise prompt into a large language model to obtain a plurality of first industry knowledge points; it can be understood that the enterprise prompt is used to guide the large language model to inquire about the first industry knowledge points related to product information in the electric sales scene, and in this step, the large language model plays the role of a purchaser by means of the enterprise prompt, and inquires about the first industry knowledge points in the electric sales scene for the goods or services sold by the enterprise.

[0062] The second knowledge point module 203 is used for knowledge condensation and knowledge deduplication of the first industry knowledge points to obtain second industry knowledge points. It can be understood that the first industry knowledge points generated by the large language model are partially cold industry knowledge that the purchasing personnel in the electric sales scene usually do not care about and consult. Taking it as a questionnaire question not only reduces the enterprise experience, but also reduces the accuracy of the answer dialogue in answering the purchasing personnel. Therefore, the first industry knowledge points also need to be knowledge condensed. When the first industry knowledge points are knowledge condensed, the large language model is inquired multiple times to obtain a plurality of first industry knowledge points. The plurality of first industry knowledge points are cross-matched using a semantic matching model, and high-frequency first industry knowledge points are screened out through a voting mechanism. The definition of high frequency can be set in advance to be that the number of occurrences or the proportion of the first industry knowledge point reaches a certain threshold before extraction. Through knowledge condensation, the dialogue questionnaire is more focused on industry hot knowledge, ensuring that the dialogue made is more in line with the actual concerns of the purchasing personnel.

[0063] Since the first industry knowledge points contain knowledge points with similar or identical meanings but different expressions, the first industry knowledge points also need to be knowledge deduplicated. When the first industry knowledge points are knowledge deduplicated, the knowledge condensed first industry knowledge points are repeatedly determined by a semantic matching model to remove the same knowledge points in the first industry knowledge points.

[0064] The questionnaire generation module 204 is used for constructing a questionnaire prompt according to the second industry knowledge points, inputting the questionnaire prompt into the large language model to obtain a dialogue questionnaire, and obtaining enterprise knowledge through the dialogue questionnaire. Specifically, the questionnaire prompt is used to guide the large language model to convert the second industry knowledge points into questions and reference answer elements, generate reference answer samples based on the reference answer elements, and generate dialogue questionnaires in combination with the questions, reference answer elements, and reference answer samples. In the questionnaire prompt, the knowledge points are mainly converted into corresponding knowledge point questions based on the role of the third-party consultant. Further, the enterprise provides enterprise knowledge to the large language model by answering the dialogue questionnaire made by the large language model, which is used to construct accurate reply dialogues. Further, by providing reference answer elements and reference answer samples respectively, the enterprise can help the enterprise to quickly generate reply dialogues for different questions.

[0065] The script generation module 205 is used to construct a script prompt based on enterprise knowledge. This prompt is then input into the large language model to obtain the corresponding response script. Understandably, because the acquired enterprise knowledge lacks accuracy and clarity in tone and format, it is insufficient to meet the requirements of a response script for a purchasing agent. Therefore, the large language model still needs to organize the enterprise knowledge to obtain formal response content. Specifically, the script prompt guides the large language model to transform enterprise knowledge into response scripts based on the customer service role.

[0066] The dialogue script construction module 206 is used to acquire the voice response information of the response script and construct a voice Q&A dialogue script library, which includes questions and corresponding voice response information. In an optional embodiment, after performing knowledge condensation and deduplication on the first industry knowledge point to obtain the second industry knowledge point, the module further includes: constructing a question prompt based on the second industry knowledge point, the question prompt being used to guide the large language model to acquire several question formats corresponding to the second industry knowledge point; and supplementing the several question formats of the second industry knowledge point into the voice Q&A dialogue script library. Specifically, through the above scheme, the common industry questions asked by buyers corresponding to each knowledge point are enriched, further improving the reliability of the voice Q&A dialogue script library and avoiding the problem that the voice Q&A dialogue script library cannot call the corresponding voice response information when faced with different questions from buyers regarding the same knowledge point.

[0067] The above embodiments are deployed based on multiple servers and nodes, enabling the device to support the large language model service capabilities built by switching between multiple accounts, ensuring automatic switching when a single large language model, server, or node fails, and stably guaranteeing the application of the large language model.

[0068] For a detailed description of the speech question answering construction device based on the large language model, please refer to the detailed description of the speech question answering construction method based on the large language model, which will not be repeated here.

[0069] Example 3

[0070] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a computer storage medium disclosed in an embodiment of the present invention. Figure 3 As shown, this embodiment of the invention discloses a computer storage medium 301, which stores computer instructions. When these computer instructions are invoked, they are used to execute the steps in the speech question-answering construction method based on a large language model disclosed in Embodiment 1 of the invention.

[0071] The apparatus embodiments described above are only illustrative, wherein the modules illustrated as separate components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules, i.e., can be located in one place or distributed to multiple network modules. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0072] Through the specific description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, including a Read-Only Memory (ROM), a Random Access Memory (RAM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), a One-time Programmable Read-Only Memory (OTPROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM) or other optical disk storage, a magnetic disk storage, a magnetic tape storage, or any other computer readable medium that can be used to carry or store data.

[0073] Finally, it should be noted that: the large language model-based voice question and answer construction method, device and medium disclosed by the embodiments of the present application are only the preferred embodiments of the present application, and are used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that; it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for constructing a speech question-answering system based on a large language model, characterized in that, The method includes: Enterprise information is obtained through voice or text interaction, including the enterprise name, script name, and product information; A company prompt is constructed based on the company information. The company prompt is then input into a large language model to obtain several first industry knowledge points. The company prompt is used to guide the large language model to query first industry knowledge points related to the product information in a telemarketing scenario. The knowledge points of the first industry are refined and deduplicated to obtain the knowledge points of the second industry. Based on the second industry knowledge point, a questionnaire prompt is constructed. The questionnaire prompt is input into a large language model to obtain a conversational questionnaire. Enterprise knowledge is obtained through the conversational questionnaire. The questionnaire prompt is used to guide the large language model to transform the second industry knowledge point into questions and reference answer elements. Reference answer examples are generated based on the reference answer elements. The conversational questionnaire is generated by combining the questions, reference answer elements and reference answer examples. A prompt is constructed based on the enterprise knowledge, and the prompt is input into a large language model to obtain response scripts. The prompt is used to guide the large language model to transform the enterprise knowledge into response scripts based on customer service roles. The voice response information of the response script is obtained, and a voice question-and-answer script library is constructed, which includes questions and corresponding voice response information.

2. The speech question-answering construction method based on a large language model according to claim 1, characterized in that, When condensing knowledge points in the first industry, a semantic matching model is used to cross-match several knowledge points in the first industry, and a voting mechanism is used to select the most frequent knowledge points in the first industry.

3. The speech question-answering construction method based on a large language model according to claim 2, characterized in that, When performing knowledge deduplication on the first industry knowledge points, the semantic matching model is used to determine duplicates of the first industry knowledge points after knowledge condensation in order to remove identical knowledge points from the first industry knowledge points.

4. The speech question-answering construction method based on a large language model according to claim 2, characterized in that, When the voice question-and-answer dialogue database is running, knowledge discovery is performed at a set period. The knowledge discovery includes: collecting questions and clustering them through the semantic matching model to obtain third industry knowledge points, and supplementing the first industry knowledge points with the third industry knowledge points.

5. The speech question-answering construction method based on a large language model according to claim 1, characterized in that, After extracting and deduplicating the knowledge points from the first industry and obtaining the knowledge points from the second industry, the process also includes: Based on the second industry knowledge point, a question prompt is constructed. The question prompt is used to guide the large language model to obtain several question formats corresponding to the second industry knowledge point. Several question formats related to the second industry knowledge point will be added to the voice Q&A script library.

6. The speech question-answering construction method based on a large language model according to claim 1, characterized in that, When obtaining the voice response information of the response script, the text-based response script is converted into the voice response information through the TTS interface.

7. A speech question-answering construction device based on a large language model, characterized in that, The device includes: The information acquisition module is used to acquire enterprise information through voice or text interaction, including enterprise name, script name and product information. The first knowledge point module is used to construct an enterprise prompt based on the enterprise information, and input the enterprise prompt into the large language model to obtain several first industry knowledge points. The enterprise prompt is used to guide the large language model to query the first industry knowledge points related to the product information in the telemarketing scenario. The second knowledge point module is used to extract and deduplicat knowledge from the first industry knowledge point in order to obtain the second industry knowledge point. The questionnaire generation module is used to construct a questionnaire prompt based on the second industry knowledge point, input the questionnaire prompt into a large language model to obtain a scripted questionnaire, and obtain enterprise knowledge through the scripted questionnaire. The questionnaire prompt is used to guide the large language model to transform the second industry knowledge point into questions and reference answer elements, generate reference answer examples based on the reference answer elements, and generate a scripted questionnaire by combining the questions, reference answer elements, and reference answer examples. The script generation module is used to construct a script prompt based on the enterprise knowledge, and input the script prompt into a large language model to obtain response scripts. The script prompt is used to guide the large language model to transform the enterprise knowledge into response scripts based on customer service roles. The dialogue construction module is used to obtain the voice response information of the response dialogue and construct a voice question and answer dialogue library, which includes questions and corresponding voice response information.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the speech question-answering construction method based on a large language model as described in any one of claims 1-6.

Images