Man-machine conversation method and device, storage medium and program product

Abstract

The invention provides a man-machine conversation method and device, a storage medium and a program product. The method comprises the steps that whether an input problem relates to at least one target professional knowledge point or not is detected; when the input question relates to at least one target professional knowledge point, constructing a large model reading understanding cue word according to each target professional knowledge point and the input question; and inputting the large model reading understanding cue word into the large language model to determine a corresponding reply answer. Therefore, through an intelligent target knowledge point detection and cue word construction mechanism, the dialogue system can automatically adjust the answer strategy according to questions in different fields, and more accurate and professional answers can be provided for users.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a human-computer dialogue method, device, storage medium and program product. Background Art

[0002] With the rapid development of artificial intelligence technology, intelligent dialogue systems have been widely used in many fields, such as customer service, intelligent assistants, education and training, etc. At present, large language models have made remarkable achievements in the field of natural language processing. Some human-computer interaction products have greatly improved the intelligence level of human-computer interaction products by integrating large language models.

[0003] Although the dialogue system based on the big model can effectively support a variety of user questions, the big model may not be accurate enough in applying professional knowledge in specific fields, and there are also limitations in knowledge customization, making it difficult to meet the needs of specialized human-computer interaction such as human-computer dialogue based on internal corporate knowledge or other specific scenarios.

[0004] The industry has not yet proposed a better solution to the above problems. Summary of the invention

[0005] The present application provides a human-computer dialogue method, device, storage medium and program product to at least solve the problem that existing large model-based dialogue systems lack accuracy when facing professional field problems.

[0006] In a first aspect, an embodiment of the present application provides a human-computer dialogue method, including: detecting whether an input question involves at least one target professional knowledge point; when the input question involves at least one target professional knowledge point, constructing a large model reading comprehension prompt word according to each of the target professional knowledge points and the input question; inputting the large model reading comprehension prompt word into a large language model to determine a corresponding reply answer.

[0007] In a second aspect, an embodiment of the present application provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the human-computer dialogue method of any embodiment of the present application.

[0008] In a third aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the human-computer dialogue method of any embodiment of the present application are implemented.

[0009] In a fourth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the human-computer dialogue method of any embodiment of the present application.

[0010] The beneficial effects of the embodiments of the present application are: By detecting whether the input question involves at least one target professional knowledge point, it can intelligently identify whether the user's question belongs to a specific field or professional category. If the question involves a professional field, the system will construct the corresponding large model reading comprehension prompt words based on the target knowledge points in the field, thereby enhancing the large language model's understanding and application accuracy of professional field knowledge. Therefore, through the intelligent target knowledge point detection and prompt word construction mechanism, the dialogue system can automatically adjust its answer strategy according to questions in different fields, and effectively overcome the challenge of lack of precision of the dialogue system based on the large model when facing professional field problems, and can provide users with more accurate and professional answers. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0012] Figure 1 An operation flow chart of an example of a human-computer dialogue method according to an embodiment of the present application is shown; Figure 2 An operation flow chart showing an example of detecting target professional knowledge points involved in an input question according to an embodiment of the present application; Figure 3 An operation flow chart showing another example of detecting target professional knowledge points involved in an input question according to an embodiment of the present application; Figure 4 An operation flow chart of an example of a human-computer dialogue method according to an embodiment of the present application is shown; Figure 5 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are 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 creative work are within the scope of protection of this application.

[0014] It should be noted that the current development of dialogue systems for a single business field still mainly relies on specific business field knowledge. The dialogue system built based on specific business field knowledge can effectively support human-computer dialogue in a single business field and is conducive to developer maintenance. However, when faced with user questions outside the specific business field, the performance of the intelligent dialogue system will drop sharply.

[0015] A single dialogue system greatly reduces system complexity and facilitates optimization and iteration, but it is not user-friendly. Generally speaking, in a known single business field, such as a business scenario that only targets document question and answer, a single dialogue system based on RAG (Retrieval-Augmented Generation) and a large model can effectively meet most user needs. At the same time, it is more convenient for developers to solve some bad cases, so developers prefer to use a single dialogue system to serve certain specific users. However, this approach is very user-unfriendly. When users ask questions outside the scenario, the dialogue experience of the single dialogue system will drop sharply.

[0016] In actual business scenarios, user questions are rich and varied. Sometimes they can be solved by relying on the reasoning ability of the large model itself (such as asking mathematical questions). However, sometimes they require consulting document knowledge to respond (such as asking about company rules and regulations) or calling specific plug-ins to respond (such as the OA plug-in of the company system). A single dialogue system cannot cover the rich and diverse dialogue scenarios.

[0017] It should be noted that the introduction to the current related technologies described in this article is only intended to help readers more clearly understand the inventive spirit of this application. It does not belong to the prior art and may be a technology that is currently being developed but not disclosed.

[0018] In some current related technologies, human-computer dialogue technologies based on large models mostly classify questions in advance. That is, before user questions are solved using a single dialogue system, a pre-classification model can be used to distinguish dialogue scenarios for user questions.

[0019] However, in actual business scenarios, when a single dialogue system is used to serve users, the system cannot limit the content of questions asked by end-side users. Therefore, when users ask questions outside the single dialogue system, it is often impossible to effectively respond to user questions.

[0020] For example, a single dialogue system based on plug-ins and large models and a single dialogue system based on RAG and large models are deployed. When the system receives a user question, it uses a pre-classification method to first classify the user question, and then calls the corresponding single dialogue system to answer the user question based on the classification result. However, based on the pre-classification step, it will increase the time consumption of the entire dialogue link and there will be a probability problem of misclassification.

[0021] In view of this, Figure 1 An operation flow chart of an example of a human-computer dialogue method according to an embodiment of the present application is shown.

[0022] like Figure 1 As shown, in step S110, it is detected whether the input question involves at least one target professional knowledge point.

[0023] In some embodiments, by performing semantic analysis on the questions input by the user, for example, through NLP (Natural Language Processing) technology, keywords, professional terms and possible industry terms in the input questions are extracted to identify whether they are related to one or more professional knowledge points. This can be achieved through keyword semantic analysis, knowledge base retrieval or plug-in analysis, etc., which is not limited here.

[0024] It should be understood that the scope or type of professional knowledge points can be diverse, such as knowledge points in proprietary fields such as medicine, law, and finance. In addition, it can also be knowledge within a specific scope specified by the user, such as corporate rules and regulations.

[0025] In step S120, when the input question involves at least one target professional knowledge point, a large model of reading comprehension prompt words is constructed according to each target professional knowledge point and the input question.

[0026] Here, by fusing the input question with the professional knowledge points, the reading comprehension prompt words required by the large model are automatically constructed to ensure that the prompt words can accurately guide the large model to focus on the impact of the professional knowledge points on the question. By optimizing the structure of the prompt words, it can most effectively guide the large model to answer.

[0027] In step S130, the large model reading comprehension prompt words are input into the large language model to determine the corresponding reply answer.

[0028] It should be understood that the types of large language models can be diverse, such as the GPT series, Qwen large model, etc., and there should be no limitation here.

[0029] Here, the large language model not only receives the original question input by the user, but also receives the related professional knowledge points. Based on the large language model's powerful language understanding and contextual reasoning capabilities, combined with the target professional knowledge points, it generates appropriate answers to ensure that the answers are highly professional and accurate.

[0030] Through the embodiments of the present application, by combining professional knowledge point detection with large model prompt word generation, it can not only support dialogue needs in multiple fields, but also provide customized answers to professional questions in different fields, improve the professionalism and accuracy of its responses to questions in specific fields, and meet the growing diversified and professional question and answer needs.

[0031] Figure 2 An operational flowchart of an example of detecting target professional knowledge points involved in an input question according to an embodiment of the present application is shown.

[0032] like Figure 2 As shown, in step S210, text matching is performed based on the input question and the preset background knowledge document to determine whether there is at least one matching target background knowledge.

[0033] Here, the background knowledge document stores at least one piece of background knowledge. The content storage form of the background knowledge document can be diverse, and it can be a structured document (such as a database, a knowledge graph) or an unstructured document (such as a natural language text). For example, it can contain multiple domain expertise, and each domain expertise is associated with a corresponding index or knowledge point serial number.

[0034] In addition, the algorithm formula for text matching can also be diversified, such as performing semantic matching or vector distance calculation between the keywords in the question and the various background knowledge in the background knowledge document, so as to identify the matching target background knowledge.

[0035] In step S220, when it is determined that there is at least one matching target background knowledge, each target background knowledge is determined as a corresponding first target professional knowledge point.

[0036] Exemplarily, once the text matching algorithm determines that there is background knowledge with a high degree of match between the input question and the background knowledge document, the system can annotate these knowledge points or call the knowledge points through the index. On the other hand, when the text matching result indicates that there is no background knowledge with a high degree of match in the background knowledge document, it is determined that there is no corresponding first target professional knowledge point.

[0037] Figure 3An operational flowchart of another example of detecting target professional knowledge points involved in an input question according to an embodiment of the present application is shown.

[0038] like Figure 3 As shown, in step S310, it is determined from a preset professional plug-in list whether there is at least one target professional plug-in that matches the input question.

[0039] Here, the professional plug-in list includes multiple plug-ins and corresponding plug-in description information. These plug-ins can cover knowledge in different fields or professions, which can teach and provide functions or services for one or more specific problems (or professional problems). In addition, the plug-in description information can include the plug-in name, function description, plug-in call address and output type, etc. It should be understood that the plug-in matching method can be diverse, for example, by comparing the keywords of the input question with the plug-in description information of each plug-in for similarity, so as to match the relevant plug-ins.

[0040] In some embodiments, thanks to the excellent natural language understanding ability of the large language model, the matching of the target professional plug-in can also be achieved by calling the large language model. Exemplarily, the preset professional plug-in list and the input question are filled into the predefined plug-in judgment prompt word template to determine the plug-in judgment prompt word. Then, the plug-in judgment prompt word is input into the large language model to determine whether there is at least one target professional plug-in matching the input question in the professional plug-in list.

[0041] It should be noted that the plug-in judgment prompt word template is a predefined framework for integrating professional plug-in lists and input questions to provide the large language model with sufficient information to determine whether there is a matching plug-in. Exemplarily, the plug-in judgment prompt word template may include the following items: input question, plug-in list, and target. Input the filled plug-in judgment prompt words into the large language model (such as GPT, Qwen, etc.), and use the natural language understanding ability to let the large language model determine which plug-in best matches the input question. The matching method based on the large language model can not only understand the context and domain of the input question, but also handle more complex and ambiguous problems.

[0042] Therefore, without relying on rules or keyword matching, through predefined templates and efficient reasoning of large language models, plug-in matching can intelligently understand the context and domain of the problem, thereby automatically selecting the most relevant plug-in, which can better meet the diverse needs of users while maintaining efficient and accurate responses, greatly improving the overall user experience and system adaptability.

[0043] In step S320, when it is determined that there is at least one matching target professional plug-in, the plug-in output results of each target professional plug-in for the input question are obtained, and each plug-in output result is determined as a corresponding second target professional knowledge point.

[0044] In some implementations, each target professional plug-in is called to respond to the input question, the input question is converted into the parameters required by the plug-in, the plug-in is called and its execution result is obtained. For example, if the target professional plug-in is a disease diagnosis plug-in, the input question may be "I have a fever, what should I do?" The system will convert the question into parameters that the plug-in can understand (such as the symptom of "fever"), and then determine the output result of the plug-in as the second target professional knowledge point.

[0045] In some cases, if multiple professional plug-ins match and return results, the output results of each plug-in can be weighted according to the plug-in matching degree to obtain the final second target professional knowledge point.

[0046] Through the embodiments of the present application, by intelligently analyzing input questions and matching the most relevant plug-ins, the system can efficiently call professional plug-ins, utilize the professional knowledge points output by the plug-ins, and flexibly integrate knowledge in various professional fields to ensure the generation of high-quality answers that meet actual needs.

[0047] In some examples of the embodiments of the present application, the background knowledge document and / or the professional plug-in list may be constructed and maintained by the user.

[0048] Specifically, the document uploaded by the user is obtained, and the background knowledge document is constructed or updated based on the document uploaded by the user. Alternatively, the plug-in list uploaded by the user is obtained, and the professional plug-in list is constructed or updated based on the plug-in list uploaded by the user.

[0049] Exemplarily, through the user interface provided by the system, the user-uploaded custom documents and custom plug-in lists are received. The documents uploaded by the user may contain multiple professional knowledge points. The system can automatically identify and extract the knowledge points (such as terms, definitions, cases, conclusions, etc.) and use them to create or update background knowledge documents. For each type of plug-in in the plug-in list uploaded by the user, after verifying its authenticity and security through static analysis and dynamic analysis, it can be used to create or update the professional plug-in list.

[0050] Through the embodiments of the present application, users can upload and maintain background knowledge documents and professional plug-in lists according to their specific needs, so that the system can automatically build and update relevant knowledge bases and plug-in libraries to achieve personalized customization.

[0051] Regarding the implementation details of step S130, in some implementations, each target professional knowledge point and input question is filled into a predefined large model reading comprehension prompt word template to determine the corresponding large model reading comprehension prompt word.

[0052] Here, the large language reading comprehension prompt word template can reserve semantic slots for target professional knowledge points and input questions, and then automatically construct prompt words by filling in the semantic slots. In an example of an embodiment of the present application, corresponding semantic slots are reserved for different types of professional knowledge points, so that various types of professional knowledge points are integrated into the prompt words. In another example of an embodiment of the present application, by predefining multiple large model reading comprehension prompt word templates, for example, each template is respectively related to the professional field involved in the question (for example, medicine, finance, etc.), or each template is respectively related to the type of professional knowledge point (for example, plug-in reply, background knowledge, etc.), etc., based on the matching call of the corresponding template, the large language model can be more clearly guided to understand the context and domain characteristics of the input question, thereby helping the model to generate highly relevant and accurate answers.

[0053] In some implementations, based on the number and type of target professional knowledge points, the system will select a corresponding predefined template, fill in the target knowledge points and input questions, and construct a corresponding large model reading comprehension prompt word.

[0054] More specifically, when each target professional knowledge point only includes the first target professional knowledge point, the predefined first large model reading comprehension prompt word template is called, and each target professional knowledge point and the input question are filled into the first large model reading comprehension prompt word template to determine the corresponding large model reading comprehension prompt word. Or When each target professional knowledge point only includes the second target professional knowledge point, the predefined second largest model reading comprehension prompt word template is called, and each target professional knowledge point and the input question are filled into the second largest model reading comprehension prompt word template to determine the corresponding large model reading comprehension prompt word. Or When each target professional knowledge point includes a first target professional knowledge point and a second target professional knowledge point, a predefined third-model reading comprehension prompt word template is called, and each target professional knowledge point and input question are filled into the third-model reading comprehension prompt word template to determine the corresponding large-model reading comprehension prompt words.

[0055] In the embodiment of the present application, according to the differentiated feedback results of calling the background knowledge document and the professional plug-in list, corresponding prompt word templates are selected respectively to provide a more comprehensive and accurate answer framework, which can effectively reduce the probability of prompt word errors caused by a single prompt word template.

[0056] In some examples of the embodiments of the present application, after detecting whether the input question involves at least one target professional knowledge point, the method also includes: when the input question does not involve the target professional knowledge point, constructing a large model reading comprehension prompt word based on the input question.

[0057] Here, if the input question does not involve any specific professional knowledge (such as "What's the weather like today?" or "What color do you like?"), the large model reading comprehension prompt words are directly constructed based on the input question. Through the natural language understanding ability of the large language model, the dialogue system can still handle a variety of question types, such as common sense questions, small talk, weather inquiries, etc., without worrying about whether the user's question belongs to a specific professional field, and has strong flexibility.

[0058] It should be emphasized that, unlike the multiple dialogue systems based on pre-classification used in the current related technologies, in the embodiments of the present application, only one intelligent dialogue system is used, and the method of summarizing and splicing through multiple routes in parallel has a higher fault tolerance rate, can effectively deal with diverse and professional user problems, and has better dialogue effects. In addition, in the scenario of user-customized plug-ins or user-customized background knowledge, it is only when the user adds the corresponding content that it is determined whether the plug-in needs to be called, and RAG and plug-in calls are parallel, rather than serial links, which can shorten the time consumption and achieve more comprehensive large model input prompt information.

[0059] Figure 4 An operation flow chart of an example of a human-computer dialogue method according to an embodiment of the present application is shown.

[0060] like Figure 4 As shown, the human-computer dialogue method includes the following operations: 1) When there are pre-uploaded background knowledge documents or question-answer pairs in the system, RAG is performed first to integrate the retrieved background knowledge fragments for later use.

[0061] In some implementations, after obtaining a user's question, RAG is first performed on the provided background knowledge document and question-answer pair, and background knowledge segments that meet the conditions are retrieved according to the configured RAG strategy and are marked with serial numbers and used for later use.

[0062] If no background knowledge is provided, ignore this step.

[0063] It should be noted that RAG may or may not give a result. Since the calling cost of RAG is low, in some examples, as long as there is pre-uploaded or provided background knowledge, the RAG step will be performed by default.

[0064] 2) While RAG is in progress, it determines whether to use a plug-in and which plug-in to use based on the user's question and the currently provided plug-in list. After determining the plug-in, it calls the plug-in to obtain the result.

[0065] Here, the provided plug-in list and user questions are filled into the pre-defined plug-in judgment prompt word template. The sample template is as follows: The following list of plugins is provided to you: {{List of plugins, including plugin names and descriptions}} Based on the user's question, please determine whether you need to use a plug-in in the plug-in list and which plug-in to use.

[0066] The user's question is as follows: {{User question}} The filled prompt words are sent to the big model for reading comprehension, and the big model outputs the plug-in name.

[0067] If you do not need to use the plug-in, skip this step.

[0068] It should be noted that the large model's reading comprehension ability is used to screen the known plug-ins provided, so as to obtain the conclusion of whether to use and which plug-in to use. After the plug-in is determined, the large model's reading comprehension ability is used again to extract the parameter information contained in the user's question, and finally the plug-in is called to obtain the result. The large model has better generalization and comprehension ability, which can achieve better results when used for plug-in tool judgment and parameter extraction. In addition, higher accuracy can be achieved by performing plug-in judgment and parameter extraction separately.

[0069] 3) Fill the background knowledge fragments and plug-in output results to be used into the placeholder modules corresponding to RAG and plug-in in the prompt word template.

[0070] Here, the user question, the parameter list and description of the selected plug-in are filled into the pre-defined extraction parameter prompt word template. The sample template is as follows: The following plugins are provided to you: {{Plugin name, parameter list, description of each parameter}} Please extract the parameter values ​​that match the plugin parameter description from the user question provided.

[0071] The user question provided to you is as follows: {{User question}} The filled prompt words are sent to the big model for reading comprehension, and the big model outputs the input parameter values ​​in the plug-in.

[0072] In this way, the extracted plug-in input parameter value is input into the plug-in to obtain the result returned by the plug-in. Through the splicing of the large model reading comprehension prompt words, the information obtained in each step is unified and regularized, so that the large model can obtain more comprehensive information to achieve better response effect.

[0073] 4) Input the final prompt word into the large language model for reading comprehension and then output the answer.

[0074] Specifically, the obtained background knowledge fragments and the obtained plug-in return results are filled into the pre-defined large model reading comprehension prompt word template. The sample template is as follows: Provide you with the following background knowledge: {{Background knowledge obtained in step 1}} In the meantime, you can refer to the results obtained by external plugins: {{The plug-in return result obtained in step 4}} Please answer the user's questions based on the above content: {{User question}} In this way, the filled prompt words are input into the large language model for reading comprehension, and then the answers to the user's questions are output.

[0075] Through the embodiments of the present application, the background knowledge and the prompt words of the plug-in are spliced ​​together, which can fully combine multiple information sources in the reading comprehension process of the large language model, thereby improving the system's response quality and ability to handle complex problems, and helping to generate more accurate and comprehensive answers.

[0076] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0077] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute any of the above-mentioned human-computer interaction methods of the present application.

[0078] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes any one of the above-mentioned human-computer interaction methods.

[0079] In some embodiments, an embodiment of the present application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a human-computer dialogue method.

[0080] Figure 5 is a schematic diagram of the hardware structure of an electronic device for executing a human-computer dialogue method provided by another embodiment of the present application, such as Figure 5 As shown, the device includes: One or more processors 510 and memory 520, Figure 5 A processor 510 is taken as an example.

[0081] The device for executing the human-machine dialogue method may further include: an input device 530 and an output device 540 .

[0082] The processor 510, the memory 520, the input device 530 and the output device 540 may be connected via a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0083] The memory 520 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the human-computer dialogue method in the embodiment of the present application. The processor 510 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 520, that is, realizing the human-computer dialogue method in the above method embodiment.

[0084] The memory 520 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 520 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 520 may optionally include a memory remotely arranged relative to the processor 510, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0085] The input device 530 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 540 may include a display device such as a display screen.

[0086] The one or more modules are stored in the memory 520, and when executed by the one or more processors 510, the human-computer dialogue method in any of the above method embodiments is executed.

[0087] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0088] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to: (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0089] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0090] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0091] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

[0092] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A human-computer dialogue method, comprising: Detect whether the input question involves at least one target expertise point; When the input question involves at least one target professional knowledge point, constructing a large model reading comprehension prompt word according to each of the target professional knowledge points and the input question; The large model reading comprehension prompt words are input into the large language model to determine the corresponding reply answers.

2. The method according to claim 1, wherein: The detecting whether the input question involves at least one target professional knowledge point includes: Performing text matching based on the input question and a preset background knowledge document to determine whether there is at least one target background knowledge that matches; the background knowledge document stores at least one background knowledge; When it is determined that there is at least one matching target background knowledge, each of the target background knowledge is determined as a corresponding first target professional knowledge point.

3. The method according to claim 2, wherein: The detecting whether the input question involves at least one target professional knowledge point includes: Determining whether there is at least one target professional plug-in matching the input question from a preset professional plug-in list; the professional plug-in list includes multiple plug-ins and corresponding plug-in description information; When it is determined that there is at least one matching target professional plug-in, the plug-in output results of each of the target professional plug-ins for the input question are obtained, and each of the plug-in output results is determined as a corresponding second target professional knowledge point.

4. The method according to claim 3, further comprising: Obtaining a document uploaded by a user, and constructing or updating the background knowledge document based on the document uploaded by the user; or A plug-in list uploaded by a user is obtained, and the professional plug-in list is constructed or updated based on the plug-in list uploaded by the user.

5. The method according to claim 3, wherein: The determining whether there is at least one target professional plug-in matching the input question from a preset professional plug-in list includes: Filling a preset professional plug-in list and the input question into a predefined plug-in judgment prompt word template to determine the plug-in judgment prompt word; The plug-in judgment prompt word is input into a large language model to determine whether there is at least one target professional plug-in matching the input question from the professional plug-in list.

6. The method according to claim 3, wherein: When the input question involves at least one target professional knowledge point, constructing prompt words according to each of the target professional knowledge points and the input question includes: Fill each of the target professional knowledge points and the input question into a predefined large model reading comprehension prompt word template to determine the corresponding large model reading comprehension prompt word.

7. The method according to claim 6, wherein: Filling each of the target professional knowledge points and the input question into the large model reading comprehension prompt word template to determine the corresponding large model reading comprehension prompt words includes: When each of the target professional knowledge points only includes the first target professional knowledge point, calling a predefined first large model reading comprehension prompt word template, filling each of the target professional knowledge points and the input question into the first large model reading comprehension prompt word template to determine a corresponding large model reading comprehension prompt word; or When each of the target professional knowledge points only includes the second target professional knowledge point, calling a predefined second large model reading comprehension prompt word template, filling each of the target professional knowledge points and the input question into the second large model reading comprehension prompt word template to determine a corresponding large model reading comprehension prompt word; or When each of the target professional knowledge points includes the first target professional knowledge point and the second target professional knowledge point, the predefined third large model reading comprehension prompt word template is called, and each of the target professional knowledge points and the input question is filled into the third large model reading comprehension prompt word template to determine the corresponding large model reading comprehension prompt word.

8. The method according to any one of claims 1 to 7, wherein: After detecting whether the input question involves at least one target professional knowledge point, the method further includes: When the input question does not involve the target professional knowledge point, a large model of reading comprehension prompt words is constructed according to the input question.

9. A storage medium having a computer program stored thereon, wherein: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 8 are implemented.

10. An electronic device, comprising: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method described in any one of claims 1 to 8.

Images