Multi-round dialogue method and device of intelligent dialogue system based on large model, and medium

Abstract

The invention provides a multi-round dialogue method and device of an intelligent dialogue system based on a large model and a medium, and relates to the technical field of data processing.The method comprises the steps that according to an identifier of a current session transmitted by an upstream service system, multiple domain agents into which the current session enters are obtained, and according to the identifier of each domain agent, a multi-round dialogue corresponding to the current session is obtained; obtaining knowledge definition description information and an enhancement mode of to-be-enhanced knowledge corresponding to each domain agent; according to a request question input by a user in the current session and the knowledge definition description information of the to-be-enhanced knowledge, obtaining target enhanced knowledge corresponding to each domain agent; according to the target enhanced knowledge, using each domain agent to generate domain knowledge information of each domain agent; and according to the domain knowledge information of the plurality of domain agents, generating target knowledge information as real-time session content of the current session and returning the real-time session content to the upstream business system. The accuracy of knowledge supplement and the task completion efficiency can be improved, and the user experience is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a multi-turn dialogue method, device, and medium for an intelligent dialogue system based on a large model. Background Technology

[0002] With the rapid development of artificial intelligence technology, intelligent dialogue systems have been widely applied in many industries such as finance, e-commerce, healthcare, education, and government affairs, becoming a key tool for improving service efficiency. These systems mainly rely on natural language understanding technology to achieve accurate intent recognition and slot extraction. At the same time, they need to maintain contextual coherence in multi-turn dialogues, achieve dynamic task state tracking, and provide a highly personalized interactive experience to meet the complex business needs in different scenarios.

[0003] Currently, existing methods are mainly based on traditional natural language processing frameworks, employing a pipelined architecture that combines intent recognition and slot extraction. Intent recognition typically uses classification models to map request questions to preset intent categories, while slot extraction uses sequence labeling models to extract key structured parameters. Some systems also introduce rule engines to enhance recognition capabilities for specific scenarios.

[0004] However, existing technologies lack sufficient integration with the deep semantic understanding and generation capabilities of large language models, and are weak in the design of mechanisms such as state modeling, context integration, and dynamic updates. These systems struggle to effectively parse complex referential relationships and intent evolution, resulting in limited overall performance when dealing with complex multi-turn dialogues in real-world scenarios, severely impacting user experience and task completion efficiency. Summary of the Invention

[0005] The purpose of this application is to address the shortcomings of the prior art by providing a multi-turn dialogue method, device, and medium for an intelligent dialogue system based on a large model, so as to improve the accuracy of knowledge supplementation and task completion efficiency, and enhance the user experience.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, one embodiment of this application provides a multi-turn dialogue method for an intelligent dialogue system based on a large model, the method comprising: Based on the identifier of the current session transmitted by the upstream business system, obtain the multiple domain agents that the current session has entered, wherein the multiple domain agents are dialogue agents for multiple specific business domains; Based on the identifiers of intelligent agents in each domain, obtain the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each intelligent agent in each domain; Based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, knowledge enhancement is performed on the knowledge to be enhanced to obtain the target enhanced knowledge corresponding to the intelligent agents in each domain; Based on the target-enhanced knowledge, the domain knowledge information of each domain agent is generated using the domain agents. Based on the domain knowledge information of the multiple domain agents, target knowledge information is generated and returned to the upstream business system as the real-time session content of the current session.

[0007] Optionally, obtaining the multiple domain agents that the current session has entered based on the identifier of the current session transmitted by the upstream business system includes: Based on the identifier of the current session, query the dialogue history of the current session; Information is extracted based on the dialogue history to obtain the historical session status information of the current session; Based on the historical session state information, obtain the list of domain agent identifiers that the current session has entered, and the list of domain agent identifiers stores the identifiers of the multiple domain agents.

[0008] Optionally, the knowledge definition description information is used to indicate the knowledge type of the knowledge to be enhanced; the enhancement mode is used to indicate the enhancement method of the knowledge to be enhanced; The step of enhancing the knowledge to be enhanced based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, to obtain the target enhanced knowledge corresponding to the intelligent agents in each domain, includes: The request question is segmented into words to obtain the segmentation results; Based on the knowledge definition description information of the knowledge to be enhanced, the enhancement mode, the word segmentation result, the request question, and the historical session information, construct the enhanced knowledge recognition request corresponding to the knowledge type; The knowledge enhancement method described above is used to perform knowledge enhancement recognition processing on the knowledge enhancement recognition request to obtain the target enhanced knowledge.

[0009] Optionally, the knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is an intent knowledge type; the enhancement method is an intent knowledge enhancement method; and the enhanced knowledge recognition request is an intent recognition request. The step of using the enhancement method of the knowledge to be enhanced to perform knowledge enhancement recognition processing on the knowledge enhancement recognition request to obtain the target enhanced knowledge includes: The intent knowledge enhancement method described above is used to identify the intent recognition request and obtain the target intent knowledge as the target enhancement knowledge.

[0010] Optionally, the intent knowledge enhancement method includes: a first intent recognition method or a second intent recognition method, wherein the intent recognition request is processed using the intent knowledge enhancement method to obtain target intent knowledge as the target enhanced knowledge, including: According to the first intent recognition method, a preset first natural language processing (NLP) model is used to recognize and process the intent recognition request, and the first target intent knowledge output by the first NLP model is used as the target enhancement knowledge. Alternatively, according to the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request to obtain the corresponding intent recognition prompt words; Based on the intent recognition prompts, a preset large language model is used to perform intent recognition, and the second target intent knowledge output by the preset large language model is used as the target enhancement knowledge.

[0011] Optionally, the intent knowledge enhancement method includes: a first intent recognition method and a second intent recognition method, and the enhancement mode is further used to indicate the intent knowledge fusion strategy of the first intent recognition method and the second intent recognition method; The step of using the intent knowledge enhancement method to identify the intent recognition request and obtain target intent knowledge as the target enhancement knowledge includes: According to the first intent recognition method, a preset first natural language processing (NLP) model is used to recognize and process the intent recognition request, and the first target intent knowledge output by the first NLP model is obtained. According to the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request to obtain the corresponding intent recognition prompt words; Based on the intent recognition prompt words, a preset large language model is used to perform intent recognition, and the second target intent knowledge output by the preset large language model is obtained. The intent knowledge fusion strategy is used to fuse the first target intent knowledge and the second target intent knowledge to obtain the target enhancement knowledge.

[0012] Optionally, the knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is a slot knowledge type; the enhancement method is a slot knowledge enhancement method; and the enhanced knowledge recognition request is a slot recognition request. The step of using the enhancement method of the knowledge to be enhanced to perform knowledge enhancement recognition processing on the knowledge enhancement recognition request to obtain the target enhanced knowledge includes: The slot knowledge enhancement method is used to identify and process the slot recognition request to obtain target slot knowledge as the target enhanced knowledge.

[0013] Optionally, the slot knowledge enhancement method includes: a first slot recognition method or a second slot recognition method, wherein the step of using the slot knowledge enhancement method to perform recognition processing on the slot recognition request to obtain target slot knowledge as the target enhanced knowledge includes: Based on the first slot recognition method, a preset second NLP model is used to recognize and process the intent recognition request, and the first target slot knowledge output by the second NLP model is used as the target augmentation knowledge. Alternatively, according to the second slot recognition method, a preset slot recognition prompt word template is used to construct prompt words for the slot recognition request to obtain the corresponding slot recognition prompt words; Based on the slot identification prompts, a preset large language model is used for slot identification to obtain the second target slot knowledge output by the preset large language model as the target augmentation knowledge.

[0014] Optionally, the slot knowledge enhancement method includes: a first slot identification method and a second slot identification method, and the enhancement mode is further used to indicate the slot knowledge fusion strategy of the first slot identification method and the second slot identification method; The step of using the slot knowledge enhancement method to identify and process the slot recognition request to obtain target slot knowledge as the target enhanced knowledge includes: Based on the first slot recognition method, a preset second NLP model is used to recognize and process the slot recognition request, thereby obtaining the first target slot knowledge output by the second NLP model; According to the second slot recognition method, a preset slot recognition prompt word template is used to construct prompt words for the slot recognition request to obtain the corresponding slot recognition prompt words; Based on the slot recognition prompts, an intent recognition is performed using a preset large language model to obtain the second target slot knowledge output by the preset large language model. The target slot knowledge is obtained by fusing the first target slot knowledge and the second target slot knowledge using the slot knowledge fusion strategy.

[0015] Secondly, another embodiment of this application provides a multi-turn dialogue device for an intelligent dialogue system, the device comprising: The first acquisition module is used to acquire multiple domain agents that the current session has entered based on the identifier of the current session transmitted by the upstream business system, wherein the multiple domain agents are dialogue agents for multiple specific business domains. The second acquisition module is used to acquire the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each domain intelligent agent based on the identifier of each domain intelligent agent; The enhancement module is used to enhance the knowledge to be enhanced based on the request question input by the user in the current session and the knowledge definition description information of the knowledge to be enhanced, so as to obtain the target enhanced knowledge corresponding to the intelligent agents in each domain. The first generation module is used to generate domain knowledge information of each domain agent based on the target augmentation knowledge and using the domain agents of each domain. The second generation module is used to generate target knowledge information based on the domain knowledge information of the multiple domain intelligent agents and return it to the upstream business system as the real-time session content of the current session.

[0016] Thirdly, another embodiment of this application provides a computer device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the multi-turn dialogue method of the large-model-based intelligent dialogue system as described in any of the first aspects above.

[0017] Fourthly, another embodiment of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the multi-turn dialogue method for a large-model-based intelligent dialogue system as described in any of the first aspects above.

[0018] The beneficial effects of this application are: This application provides a multi-turn dialogue method, device, and medium for an intelligent dialogue system based on a large model. Based on the identifier of the current session transmitted from the upstream business system, it obtains multiple domain agents already involved in the current session. Based on the identifiers of each domain agent, it obtains the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each domain agent. Based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, it performs knowledge enhancement on the knowledge to be enhanced, obtaining the target enhanced knowledge corresponding to each domain agent. Based on the target enhanced knowledge, it uses each domain agent to generate domain knowledge information for each domain agent. Based on the domain knowledge information of multiple domain agents, it generates target knowledge information as the real-time conversation content of the current session and returns it to the upstream business system. This application determines the corresponding domain agent through the current session identifier, allowing different user questions to activate different agent combinations, eliminating the need for manual maintenance of complex flowcharts and adapting to rapidly changing technical field requirements. Simultaneously, by obtaining the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each domain agent, it can improve the accuracy of knowledge supplementation based on a semantically aligned knowledge enhancement mechanism, ensuring that each knowledge enhancement has a clear target and improving system learning efficiency. Furthermore, by determining the domain knowledge information of each domain's intelligent agents, target knowledge information can be obtained, enabling real-time feedback and continuous evolution of the conversation, and providing immediate, accurate, and actionable feedback. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a multi-turn dialogue method for an intelligent dialogue system based on a large model, provided in an embodiment of this application; Figure 2 A flowchart illustrating the process of determining multiple domain agents in a multi-turn dialogue method for an intelligent dialogue system based on a large model, provided in an embodiment of this application. Figure 3 A flowchart illustrating the process of determining target augmentation knowledge in a multi-turn dialogue method for an intelligent dialogue system based on a large model, provided in an embodiment of this application. Figure 4 A schematic diagram illustrating the process of determining intent recognition in another multi-turn dialogue method for an intelligent dialogue system based on a large model, provided in an embodiment of this application. Figure 5A schematic diagram illustrating the process of determining intent recognition in a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in an embodiment of this application. Figure 6 A schematic diagram illustrating the process of determining slot recognition in a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in an embodiment of this application. Figure 7 A schematic diagram illustrating the process of determining slot recognition in a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in an embodiment of this application. Figure 8 A schematic diagram of the structure of a multi-turn dialogue device for an intelligent dialogue system based on a large model, provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0022] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0023] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0024] To clearly describe the multi-turn dialogue method for an intelligent dialogue system based on a large model provided in this application, the method will be described below in conjunction with several accompanying drawings. Figure 1A flowchart illustrating a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in this application embodiment, is shown below. Figure 1 As shown, the method includes: Step 101: Based on the identifier of the current session transmitted by the upstream business system, obtain the multiple domain agents that the current session has entered.

[0025] The current session is the large model session. Multiple domain agents are dialogue agents targeting specific business domains; these domain agents can be domain bots, and the agents themselves are the agents within the large model. The upstream business system can be an application or website that needs to access intelligent dialogue capabilities. The corresponding current session identifier and the natural language request question are obtained from the upstream business system. The identifier of the current session is the current session identifier (IdentityDocument, or ID for short).

[0026] Optionally, based on the identifier of the current session transmitted from the upstream business system, it is determined whether the historical session rounds corresponding to the current session identifier are stored. If they exist, the round corresponding to the current session identifier is incremented by 1, and multiple domain agents corresponding to the historical session rounds are identified as the multiple domain agents corresponding to the current session identifier. If the identifier of the current session does not exist, a new session state is constructed, setting the corresponding identifier, time, and round, and multiple domain agents are determined based on the identifier of the current session. Specifically, the multiple domain agents can be determined based on the input request question corresponding to the identifier of the current session.

[0027] Step 102: Based on the identifiers of each domain agent, obtain the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each domain agent.

[0028] The knowledge definition description describes the knowledge type, semantic category, structural identifier, and source basis of the knowledge to be enhanced. The enhancement pattern indicates the acquisition path for each domain agent. Domain agents refer to AI entities with specific professional domains and execution capabilities. The knowledge to be enhanced is the unfilled knowledge within the domain agents.

[0029] Optionally, based on the identifiers of the domain agents, each domain agent is identified, and multiple knowledge states within each domain agent are obtained. Based on these multiple knowledge states, the knowledge to be enhanced is determined, and the knowledge definition description information and enhancement mode of the knowledge to be enhanced are obtained. The status of the multiple knowledge states indicates whether the knowledge has been filled in; if the knowledge has not been filled in, then that knowledge is the knowledge to be enhanced.

[0030] Step 103: Based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, perform knowledge enhancement on the knowledge to be enhanced to obtain the target enhanced knowledge corresponding to each domain agent.

[0031] Optionally, based on the request question input by the user in the current session and the knowledge definition description information of the knowledge to be enhanced, the content corresponding to the knowledge to be enhanced is determined from the request question, thereby realizing knowledge enhancement of the knowledge to be enhanced and obtaining the target enhanced knowledge corresponding to the intelligent agent in each domain.

[0032] Step 104: Enhance knowledge according to the goal, and generate domain knowledge information for each domain agent using agents from each domain.

[0033] Optionally, the target augmentation knowledge is added to each domain agent according to the knowledge to be augmented in each agent, and the domain knowledge information of each domain agent is generated using each domain agent.

[0034] Optionally, based on the domain knowledge information generated by each domain agent, it determines whether to trigger a new domain agent, and provides feedback on the processing confidence of each domain agent and whether to continue the session. The processing confidence is based on the current filling status of each domain agent and the accuracy of the obtained domain knowledge information.

[0035] Optionally, the processing results of each domain agent can be selected based on preset posterior scoring rules. The overall state of the entire session is refreshed by combining the selected answers and the state information of each agent. The system decides whether to retain or clear the state of the corresponding domain agent in the session storage, depending on whether the session needs to be maintained. This ensures that when subsequent user requests arrive, the system can intelligently determine whether it is necessary to re-probe the state of the domain agent to obtain the parameters required for subsequent enhanced recognition.

[0036] Step 105: Based on the domain knowledge information of multiple domain agents, generate target knowledge information as the real-time session content of the current session and return it to the upstream business system.

[0037] Optionally, target knowledge information is generated based on the domain knowledge information of multiple domain agents, and the target knowledge information is scored. Based on the scoring results, the real-time session content of the current session is generated and returned to the upstream business system. The scoring can be based on preset scoring rules, such as agent credibility, consistency of content among multiple agents, etc., which are not limited in this embodiment.

[0038] In this embodiment, based on the identifier of the current session transmitted from the upstream business system, multiple domain agents already entered in the current session are obtained. Based on the identifier of each domain agent, the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each domain agent are obtained. Based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, knowledge enhancement is performed on the knowledge to be enhanced, resulting in the target enhanced knowledge corresponding to each domain agent. Based on the target enhanced knowledge, domain knowledge information of each domain agent is generated. Based on the domain knowledge information of multiple domain agents, target knowledge information is generated and returned to the upstream business system as the real-time session content of the current session. This application determines the corresponding domain agent through the current session identifier, enabling different user queries to activate different agent combinations, eliminating the need for manual maintenance of complex flowcharts and adapting to rapidly changing technical field requirements. Simultaneously, by obtaining the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each domain agent, a semantically aligned knowledge enhancement mechanism can be used to improve the accuracy of knowledge supplementation, ensuring that each knowledge enhancement has a clear target and improving system learning efficiency. Furthermore, by determining the domain knowledge information of each domain's intelligent agents, target knowledge information can be obtained, enabling real-time feedback and continuous evolution of the conversation, and providing immediate, accurate, and actionable feedback.

[0039] Based on the above embodiments, this application also provides a process for determining multiple domain agents in a multi-turn dialogue method for a large-model-based intelligent dialogue system. Figure 2 A flowchart illustrating the process of determining multiple domain agents in a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in this application embodiment, is shown below. Figure 2 As shown, in step 101 above, based on the identifier of the current session transmitted by the upstream business system, multiple domain agents that the current session has entered are obtained, including: Step 201: Based on the identifier of the current session, query the dialogue history of the current session.

[0040] Optionally, based on the identifier of the current session, it can be determined whether the dialogue history content corresponding to the identifier of the current session is stored. If it exists, the dialogue history content corresponding to the identifier of the current session can be retrieved.

[0041] Step 202: Extract information based on the dialogue history to obtain the historical session status information of the current session.

[0042] Among them, the historical session state information consists of extracted key semantic summaries and contextual data, which are used to determine the evolution trajectory of user intent.

[0043] Optionally, information can be extracted from keywords in the historical dialogue content to obtain key semantic summaries and contextual data of the current conversation as historical conversation state information.

[0044] Step 203: Based on the historical session state information, obtain the list of domain agent identifiers that the current session has entered. The list of domain agent identifiers stores the identifiers of multiple domain agents.

[0045] Optionally, a list of corresponding domain agent identifiers is determined based on the key semantic summary and context data in the historical session state information. The list of domain agent identifiers stores the identifiers of multiple domain agents.

[0046] In this embodiment of the application, the dialogue history of the current session is queried based on the identifier of the current session; information is extracted based on the dialogue history to obtain the historical session state information of the current session; based on the historical session state information, a list of domain agent identifiers that the current session has entered is obtained, and the list of domain agent identifiers stores the identifiers of multiple domain agents.

[0047] Based on the above embodiments, the knowledge definition description information is used to indicate the knowledge type of the knowledge to be enhanced; the enhancement mode is used to indicate the enhancement method of the knowledge to be enhanced. This application also provides a process for determining target enhanced knowledge in a multi-turn dialogue method for an intelligent dialogue system based on a large model. Figure 3 A flowchart illustrating the process of determining target augmentation knowledge in a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in this application embodiment, is shown below. Figure 3 As shown, in step 103 above, based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, knowledge enhancement is performed on the knowledge to be enhanced to obtain the target enhanced knowledge corresponding to each domain agent, including: Step 301: Perform word segmentation on the request question to obtain the word segmentation result.

[0048] Among them, word segmentation is used to divide continuous sentences into independent word units.

[0049] Optionally, the request question is segmented using a preset segmentation method to obtain the segmentation result. The preset segmentation method can be a forward maximum matching method, a backward maximum matching method, or other methods; this embodiment does not limit the specific methods used.

[0050] Step 302: Based on the knowledge definition description information, enhancement mode, word segmentation results, request question and historical conversation information of the knowledge to be enhanced, construct the enhanced knowledge recognition request corresponding to the knowledge type.

[0051] The knowledge type can be determined by the knowledge definition description information, and can be either an intent type or a slot type.

[0052] Optionally, the knowledge definition description information, enhancement mode, word segmentation results, request question, and historical conversation information of the knowledge to be enhanced can be used as the enhanced knowledge recognition request corresponding to the knowledge type.

[0053] Step 303: Using the augmentation method of the knowledge to be augmented, perform knowledge augmentation recognition processing on the augmentation knowledge recognition request to obtain the target augmentation knowledge.

[0054] The enhancement methods can include Natural Language Processing (NLP), large language model processing, or a combination of natural language processing and large language model processing, etc. This application does not limit these methods.

[0055] Optionally, the augmentation method of the knowledge to be augmented is adopted to perform knowledge augmentation recognition processing on the augmentation knowledge recognition request, thereby obtaining the target augmentation knowledge.

[0056] This application constructs a structured augmented knowledge recognition request through comprehensive analysis of the knowledge definition description information of the knowledge to be augmented, the augmentation mode, the word segmentation results of the user request, the original question, and historical session state information. It then calls the matching augmentation method for processing, achieving precision, contextualization, and configurability in the knowledge augmentation process. This improves the accuracy and relevance of the target augmented knowledge acquisition and solves the technical problems of blind knowledge updates, inaccurate responses, and lack of context awareness in traditional technologies.

[0057] Based on the above embodiments, the knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is an intent knowledge type; the enhancement method is an intent knowledge enhancement method; and the enhancement knowledge recognition request is an intent recognition request. This application also provides another process for determining target enhancement knowledge in a multi-turn dialogue method of an intelligent dialogue system. In step 303 above, the enhancement method of the knowledge to be enhanced is used to perform knowledge enhancement recognition processing on the enhancement knowledge recognition request to obtain the target enhancement knowledge, including: An intent knowledge enhancement approach is adopted to identify and process intent recognition requests, thereby obtaining target intent knowledge as target enhancement knowledge.

[0058] Among them, the intent knowledge type is the semantic intent category of the request question input in the current session, the intent recognition request is the request data constructed for the current intent knowledge type, and the intent knowledge enhancement method is determined according to the current intent knowledge type. Different intent knowledge types correspond to different intent knowledge enhancement methods.

[0059] This application's embodiments process intent recognition requests using an intent knowledge enhancement approach. By combining the semantic features of user questions with historical conversation context, it accurately identifies deep user intents and generates structured, interpretable target intent knowledge. This significantly improves the accuracy and context adaptability of intent understanding, avoiding misjudgments and omissions in traditional methods. It achieves intelligent, professional, and traceable knowledge enhancement, providing a highly reliable cognitive foundation for subsequent collaborative responses of domain-specific intelligent agents.

[0060] Based on the above embodiments, the intent knowledge enhancement method includes: a first intent recognition method or a second intent recognition method. This application also provides a process for determining intent recognition in a multi-turn dialogue method of an intelligent dialogue system. In the above-mentioned intent knowledge enhancement method, the intent recognition request is identified and processed to obtain target intent knowledge as target enhanced knowledge, including: According to the first intent recognition method, a preset first natural language processing (NLP) model is used to recognize and process the intent recognition request, and the first target intent knowledge output by the first NLP model is used as target enhancement knowledge.

[0061] The first intent recognition method is NLP intent recognition. The first NLP model is a pre-trained model for intent recognition. The target augmentation knowledge may include at least one intent recognition result corresponding to the intent recognition request and the confidence level of the intent recognition result.

[0062] Optionally, based on the first intent recognition method, a preset first natural language processing (NLP) model is used to recognize and process the knowledge definition description information, enhancement patterns, word segmentation results, request questions, and historical conversation information in the intent recognition request, so as to obtain the first target intent knowledge output by the first NLP model as the target enhancement knowledge.

[0063] Based on the above embodiments, the intent knowledge enhancement method includes: a first intent recognition method or a second intent recognition method. This application also provides another process for determining intent recognition in a multi-turn dialogue method of an intelligent dialogue system based on a large model. Figure 4 A flowchart illustrating the determination of intent recognition in another multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in this application embodiment, is shown below. Figure 4 As shown, the above-mentioned intent knowledge enhancement method is used to identify and process the intent recognition request, obtaining target intent knowledge as target enhancement knowledge, including: Step 401: Based on the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request, thereby obtaining the corresponding intent recognition prompt words.

[0064] The second intent recognition method is large language model recognition. The preset intent recognition prompt word template is a pre-designed, structured text framework used to combine the information in the intent recognition request in a way that the large language model can understand.

[0065] Optionally, according to the second intent recognition method, a preset intent recognition prompt word template is used to extract keywords from the intent recognition request to construct prompt words, and the extracted keywords are filled into the preset intent recognition prompt word template to obtain the corresponding intent recognition prompt words.

[0066] Step 402: Based on the intent recognition prompts, use a preset large language model to perform intent recognition, and obtain the second target intent knowledge output by the preset large language model as target enhancement knowledge.

[0067] Optionally, based on the intent recognition prompt words, a preset large language model is used to perform intent recognition on the intent recognition request, and the second target intent knowledge output by the preset large language model is used as target enhancement knowledge.

[0068] This application provides two complementary intent recognition paths: a first approach based on a preset NLP model and a second approach based on prompt word templates and a large language model. This achieves a dual improvement in the flexibility and robustness of intent recognition. The use of a dedicated NLP model ensures the stability and accuracy of intent recognition in high-time-sensitivity scenarios; the use of prompt word templates to guide the large language model in deep semantic understanding significantly enhances the ability to analyze complex, ambiguous, or multi-layered intents. The two approaches can be switched or used in combination as needed, balancing efficiency and accuracy, effectively improving the intelligence and adaptability of target intent knowledge generation. Simultaneously, the structured prompt word design ensures the controllability and interpretability of the recognition process.

[0069] Based on the above embodiments, the intent knowledge enhancement method includes: a first intent recognition method and a second intent recognition method. The enhancement mode is further used to indicate the intent knowledge fusion strategy of the first intent recognition method and the second intent recognition method. This application also provides a process for determining intent recognition in a multi-turn dialogue method of an intelligent dialogue system based on a large model. Figure 5 A flowchart illustrating the determination of intent recognition in a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in this application embodiment, is shown below. Figure 5 As shown, the above-mentioned intent knowledge enhancement method is used to identify and process the intent recognition request, obtaining target intent knowledge as target enhancement knowledge, including: Step 501: Based on the first intent recognition method, the intent recognition request is recognized and processed using a preset first natural language processing (NLP) model to obtain the first target intent knowledge output by the first NLP model.

[0070] Step 502: Based on the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request, thereby obtaining the corresponding intent recognition prompt words.

[0071] Step 503: Based on the intent recognition prompts, use a preset large language model to perform intent recognition and obtain the second target intent knowledge output by the preset large language model.

[0072] Step 504: Employ an intent knowledge fusion strategy to fuse the first target intent knowledge and the second target intent knowledge to obtain target enhancement knowledge.

[0073] Among them, the intent knowledge fusion strategy is used to fuse the first target intent knowledge and the second target intent knowledge to obtain the accuracy of the target enhancement knowledge.

[0074] For example, the first target intent knowledge and the second target intent knowledge are scored to obtain the scoring results of the first target intent knowledge and the second target intent. If the score of the first target intent knowledge is high, according to the threshold of the fusion strategy and the intent score, the target enhancement knowledge in the first target intent knowledge that exceeds the strategy threshold is identified, and the identification method of the target enhancement knowledge is marked as the first intent identification method.

[0075] For example, if the second target intent knowledge yields a result, but the first target intent knowledge does not, then the second target intent knowledge is treated as target enhancement knowledge, and the recognition method of the target enhancement knowledge is marked as the second intent recognition method. If the second target intent knowledge is among the first target intent knowledge exceeding a threshold, then the second target intent knowledge is filtered out, and the recognition result of the first intent recognition method is used.

[0076] In this embodiment, by combining a first intent recognition method based on a preset NLP model with a second intent recognition method based on prompt word templates and a large language model, and employing an intent knowledge fusion strategy to collaboratively fuse the target intent knowledge output by both methods, this approach combines the high accuracy and low latency advantages of dedicated models with the strong semantic understanding and contextual reasoning capabilities of large language models. This effectively improves the comprehensiveness, robustness, and interpretability of intent recognition. This method avoids misjudgments or information gaps caused by a single model, and ensures the stability and professionalism of the results through structured prompts and fusion decision-making mechanisms, significantly enhancing the system's cognitive ability in complex, ambiguous, or polysemous scenarios.

[0077] Based on the above embodiments, the knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is a slot knowledge type; the enhancement method is a slot knowledge enhancement method; and the enhanced knowledge recognition request is a slot recognition request. This application also provides another process for determining target enhancement recognition in a multi-turn dialogue method of an intelligent dialogue system. In step 303 above, the enhancement method of the knowledge to be enhanced is used to perform knowledge enhancement recognition processing on the enhanced knowledge recognition request to obtain the target enhanced knowledge, including: A slot knowledge enhancement approach is adopted to identify and process slot recognition requests, and obtain target slot knowledge as target enhancement knowledge.

[0078] Among them, slot knowledge type is the key information fragment or entity identified and extracted from the request question, slot identification request is the request data constructed by the current request question for the current slot knowledge type, and slot knowledge enhancement method is determined according to the current slot knowledge type. Different slot knowledge types correspond to different slot knowledge enhancement methods.

[0079] In this embodiment, the slot knowledge enhancement method is used to process the slot recognition request. This method combines domain knowledge base, contextual semantics and structured constraint rules to accurately identify and fill in key parameters in the user's intent, significantly improving the accuracy and completeness of slot extraction.

[0080] Based on the above embodiments, the slot knowledge enhancement method includes: a first slot recognition method or a second slot recognition method. This application also provides a process for determining slot recognition in a multi-turn dialogue method of an intelligent dialogue system. Using the above-mentioned slot knowledge enhancement method, the slot recognition request is identified and processed to obtain target slot knowledge as target enhancement knowledge, including: Based on the first slot recognition method, a preset second NLP model is used to recognize and process the intent recognition request, and the first target slot knowledge output by the second NLP model is used as target augmentation knowledge.

[0081] The first slot identification method is NLP slot identification. The second NLP model is a pre-trained model for slot identification. The target augmentation knowledge may include at least one slot identification result corresponding to the slot identification request and the confidence level of the slot identification result.

[0082] Optionally, based on the first slot recognition method, a preset second NLP model is used to recognize and process the knowledge definition description information, enhancement mode, word segmentation results, request questions and historical conversation information in the slot recognition request, so as to obtain the first target slot knowledge output by the second NLP model as the target enhancement knowledge.

[0083] Based on the above embodiments, this application also provides another process for determining slot recognition in a multi-turn dialogue method for a large-model-based intelligent dialogue system. Figure 6 A flowchart illustrating the process of determining slot recognition in another multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in this application embodiment, is shown below. Figure 6 As shown, the above-mentioned slot knowledge enhancement method is used to process the slot recognition request and obtain the target slot knowledge as the target enhancement knowledge, including: Step 601: Based on the second slot recognition method, a preset slot recognition prompt word template is used to construct prompt words for the slot recognition request, thereby obtaining the corresponding slot recognition prompt words.

[0084] The second slot recognition method is large language model recognition. The preset slot recognition prompt word template is a pre-designed, structured text framework used to combine the information in the slot recognition request in a way that the large language model can understand.

[0085] Optionally, according to the second slot recognition method, a preset slot recognition prompt word template is used to extract keywords from the slot recognition request to construct prompt words, and the extracted keywords are filled into the preset slot recognition prompt word template to obtain the corresponding slot recognition prompt words.

[0086] Step 602: Based on the slot recognition prompts, use a preset large language model to perform slot recognition, and obtain the second target slot knowledge output by the preset large language model as target augmentation knowledge.

[0087] Optionally, based on the slot recognition prompt, a preset large language model is used to perform slot recognition on the slot recognition request, and the second target slot knowledge output by the preset large language model is used as target augmentation knowledge.

[0088] In this embodiment, by providing two complementary slot recognition paths, the ability to mine implicit information in complex sentences, ambiguous expressions and context-dependent scenarios is improved, effectively enhancing the robustness and flexibility of the system under diverse inputs. While ensuring the accuracy of slot recognition, the depth of semantic understanding is also taken into account, significantly improving the completeness, accuracy and interpretability of target slot knowledge generation.

[0089] Based on the above embodiments, the slot knowledge enhancement method includes: a first slot recognition method and a second slot recognition method. The enhancement mode is further used to indicate the slot knowledge fusion strategy of the first and second slot recognition methods. This application also provides another process for determining slot recognition in a multi-turn dialogue method of a large-model-based intelligent dialogue system. Figure 7 A flowchart illustrating the process of determining slot recognition in a multi-turn dialogue method for an intelligent dialogue system based on a large model, as provided in this application embodiment, is shown below. Figure 7 As shown, the above-mentioned slot knowledge enhancement method is used to process the slot recognition request and obtain the target slot knowledge as the target enhancement knowledge, including: Step 701: Based on the first slot recognition method, a preset second NLP model is used to recognize and process the slot recognition request, and the first target slot knowledge output by the second NLP model is obtained.

[0090] Step 702: Based on the second slot recognition method, a preset slot recognition prompt word template is used to construct prompt words for the slot recognition request, thereby obtaining the corresponding slot recognition prompt words.

[0091] Step 703: Based on the slot recognition prompts, use a preset large language model to perform intent recognition and obtain the second target slot knowledge output by the preset large language model.

[0092] Step 704: Employ a slot knowledge fusion strategy to fuse the first target slot knowledge and the second target slot knowledge to obtain target enhancement knowledge.

[0093] Among them, the slot knowledge fusion strategy is used to determine the target slot knowledge from the first target slot knowledge and the second target slot knowledge.

[0094] Optionally, a slot knowledge fusion strategy is employed to determine the confidence levels of the first target slot knowledge and the second target slot knowledge. The confidence levels of the first and second target slot knowledge are then fused to obtain target-enhanced knowledge. Specifically, if the confidence level of the first target slot knowledge is low, the second target slot knowledge is used as the target-enhanced knowledge; if the confidence level of the second target slot knowledge is low, the first target slot knowledge is used as the target-enhanced knowledge; if the confidence levels of both the first and second target slot knowledge are low, the second target slot knowledge is used as the target-enhanced knowledge.

[0095] This application embodiment integrates a first slot recognition method based on a preset second NLP model with a second recognition method based on a large language model guided by prompt word templates, and uses a slot knowledge fusion strategy to comprehensively process the results of the two methods. This combines the high accuracy of the dedicated model in structured slot extraction with the advantages of the large language model in semantic understanding, contextual reasoning and implicit information mining, thereby enhancing the system's ability to capture technical features in a refined manner and generating high-quality target-enhanced knowledge.

[0096] Based on the same inventive concept, this application also provides a multi-turn dialogue device for a large-model-based intelligent dialogue system, corresponding to the multi-turn dialogue method of the large-model-based intelligent dialogue system. Since the principle of the device in this application is similar to the multi-turn dialogue method of the large-model-based intelligent dialogue system described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0097] Figure 8 A schematic diagram of the structure of a multi-turn dialogue device for an intelligent dialogue system based on a large model, provided in an embodiment of this application, is shown below. Figure 8 As shown, the device includes: The first acquisition module 801 is used to acquire multiple domain agents that the current session has entered based on the identifier of the current session transmitted by the upstream business system. The multiple domain agents are dialogue agents for multiple specific business domains. The second acquisition module 802 is used to acquire the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each domain intelligent agent based on the identifier of each domain intelligent agent; The enhancement module 803 is used to enhance the knowledge to be enhanced based on the request question input by the user in the current session and the knowledge definition description information of the knowledge to be enhanced, so as to obtain the target enhanced knowledge corresponding to the intelligent agent in each domain. The first generation module 804 is used to generate domain knowledge information of each domain agent by using agents in each domain based on the target augmentation knowledge. The second generation module 805 is used to generate target knowledge information based on the domain knowledge information of multiple domain intelligent agents and return it as the real-time session content of the current session to the upstream business system.

[0098] In one possible implementation, the first acquisition module 801 is specifically used to: query the dialogue history content of the current session based on the identifier of the current session; Information is extracted from the dialogue history to obtain the historical session status information of the current session. Based on historical session state information, obtain the list of domain agent identifiers that the current session has entered. The list of domain agent identifiers stores the identifiers of multiple domain agents.

[0099] In one possible implementation, the knowledge definition description information is used to indicate the knowledge type of the knowledge to be enhanced; the enhancement mode is used to indicate the enhancement method of the knowledge to be enhanced; the enhancement module 803 is specifically used to: perform word segmentation processing on the request question to obtain the word segmentation result; Based on the knowledge definition description information of the knowledge to be enhanced, the enhancement mode, the word segmentation results, the request question and the historical conversation information, construct the enhanced knowledge recognition request corresponding to the knowledge type; By employing the augmentation method of the knowledge to be augmented, the augmentation knowledge identification request is processed to obtain the target augmentation knowledge.

[0100] In one possible implementation, the knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is an intent knowledge type; the enhancement method is an intent knowledge enhancement method, and the enhanced knowledge recognition request is an intent recognition request; the enhancement module 803 is specifically used to: use the intent knowledge enhancement method to recognize and process the intent recognition request, and obtain the target intent knowledge as the target enhanced knowledge.

[0101] In one possible implementation, the intent knowledge enhancement method includes: a first intent recognition method or a second intent recognition method. The enhancement module 803 is specifically used to: according to the first intent recognition method, use a preset first natural language processing (NLP) model to recognize and process the intent recognition request, and obtain the first target intent knowledge output by the first NLP model as the target enhancement knowledge. Alternatively, based on the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request, thereby obtaining the corresponding intent recognition prompt words; Based on the intent recognition prompts, a pre-set large language model is used for intent recognition, and the second target intent knowledge output by the pre-set large language model is used as target enhancement knowledge.

[0102] In one possible implementation, the intent knowledge enhancement method includes: a first intent recognition method and a second intent recognition method. The enhancement mode is also used to indicate the intent knowledge fusion strategy of the first intent recognition method and the second intent recognition method. The enhancement module 803 is specifically used to: perform intent recognition processing on the intent recognition request using a preset first natural language processing (NLP) model according to the first intent recognition method, and obtain the first target intent knowledge output by the first NLP model. According to the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request, and the corresponding intent recognition prompt words are obtained. Based on the intent recognition prompts, a pre-set large language model is used to perform intent recognition, and the second target intent knowledge output by the pre-set large language model is obtained. An intent knowledge fusion strategy is adopted to fuse the intent knowledge of the first target and the intent knowledge of the second target to obtain target enhancement knowledge.

[0103] In one possible implementation, the knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is a slot knowledge type; the enhancement method is a slot knowledge enhancement method, and the enhanced knowledge recognition request is a slot recognition request; the enhancement module 803 is specifically used for: A slot knowledge enhancement approach is adopted to identify and process slot recognition requests, and obtain target slot knowledge as target enhancement knowledge.

[0104] In one possible implementation, the slot knowledge enhancement method includes: a first slot recognition method or a second slot recognition method. The enhancement module 803 is specifically used to: according to the first slot recognition method, use a preset second NLP model to perform recognition processing on the intent recognition request, and obtain the first target slot knowledge output by the second NLP model as the target enhancement knowledge. Alternatively, based on the second slot recognition method, a preset slot recognition prompt word template is used to construct prompt words for the slot recognition request, thereby obtaining the corresponding slot recognition prompt words; Based on the slot recognition prompts, a pre-set large language model is used for slot recognition, and the second target slot knowledge output by the pre-set large language model is used as the target augmentation knowledge.

[0105] In one possible implementation, the slot knowledge enhancement method includes: a first slot identification method and a second slot identification method, and the enhancement mode is further used to indicate the slot knowledge fusion strategy of the first slot identification method and the second slot identification method; the enhancement module 803 is specifically used for: Based on the first slot recognition method, a preset second NLP model is used to recognize and process the slot recognition request, and the first target slot knowledge output by the second NLP model is obtained. According to the second slot recognition method, a preset slot recognition prompt word template is used to construct prompt words for the slot recognition request, and the corresponding slot recognition prompt words are obtained. Based on the slot recognition prompts, a pre-set large language model is used for intent recognition to obtain the second target slot knowledge output by the pre-set large language model. A slot knowledge fusion strategy is adopted to fuse the slot knowledge of the first target and the slot knowledge of the second target to obtain target-enhanced knowledge.

[0106] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0107] This application also provides a computer device. Figure 9 This application provides a schematic diagram of the structure of a computer device, as shown in the embodiment of the present application. Figure 9As shown, the computer device includes a processor 901 and a memory 902, and optionally, a bus 903. The memory 902 stores machine-readable instructions executable by the processor 901. When the computer device is running, the processor 901 and the memory 902 communicate via the bus 903. When the machine-readable instructions are executed by the processor 901, the steps of the multi-turn dialogue method of the large-model-based intelligent dialogue system described above are performed.

[0108] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the multi-turn dialogue method of the intelligent dialogue system described above.

[0109] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0110] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0111] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-turn dialogue method for an intelligent dialogue system based on a large model, characterized in that, The method includes: Based on the identifier of the current session transmitted by the upstream business system, obtain the multiple domain agents that the current session has entered, wherein the multiple domain agents are dialogue agents for multiple specific business domains; Based on the identifiers of intelligent agents in each domain, obtain the knowledge definition description information and enhancement mode of the knowledge to be enhanced corresponding to each intelligent agent in each domain; Based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, knowledge enhancement is performed on the knowledge to be enhanced to obtain the target enhanced knowledge corresponding to the intelligent agents in each domain; Based on the target-enhanced knowledge, the domain knowledge information of each domain agent is generated using the domain agents. Based on the domain knowledge information of the multiple domain agents, target knowledge information is generated and returned to the upstream business system as the real-time session content of the current session.

2. The method according to claim 1, characterized in that, The step of obtaining multiple domain agents that the current session has entered based on the identifier of the current session transmitted by the upstream business system includes: Based on the identifier of the current session, query the dialogue history of the current session; Information is extracted based on the dialogue history to obtain the historical session status information of the current session; Based on the historical session state information, obtain the list of domain agent identifiers that the current session has entered, and the list of domain agent identifiers stores the identifiers of the multiple domain agents.

3. The method according to claim 2, characterized in that, The knowledge definition description information is used to indicate the knowledge type of the knowledge to be enhanced; the enhancement mode is used to indicate the enhancement method of the knowledge to be enhanced; The step of enhancing the knowledge to be enhanced based on the user's request question input in the current session and the knowledge definition description information of the knowledge to be enhanced, to obtain the target enhanced knowledge corresponding to the intelligent agents in each domain, includes: The request question is segmented into words to obtain the segmentation results; Based on the knowledge definition description information of the knowledge to be enhanced, the enhancement mode, the word segmentation result, the request question, and the historical session information, construct the enhanced knowledge recognition request corresponding to the knowledge type; The knowledge enhancement method described above is used to perform knowledge enhancement recognition processing on the knowledge enhancement recognition request to obtain the target enhanced knowledge.

4. The method according to claim 3, characterized in that, The knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is an intentional knowledge type; The enhancement method is an intent knowledge enhancement method, and the enhanced knowledge recognition request is an intent recognition request; The step of using the enhancement method of the knowledge to be enhanced to perform knowledge enhancement recognition processing on the knowledge enhancement recognition request to obtain the target enhanced knowledge includes: The intent knowledge enhancement method described above is used to identify the intent recognition request and obtain the target intent knowledge as the target enhancement knowledge.

5. The method according to claim 4, characterized in that, The intent knowledge enhancement method includes: a first intent recognition method or a second intent recognition method. The step of using the intent knowledge enhancement method to recognize and process the intent recognition request to obtain target intent knowledge as the target enhanced knowledge includes: According to the first intent recognition method, a preset first natural language processing (NLP) model is used to recognize and process the intent recognition request, and the first target intent knowledge output by the first NLP model is used as the target enhancement knowledge. Alternatively, according to the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request to obtain the corresponding intent recognition prompt words; Based on the intent recognition prompts, a preset large language model is used to perform intent recognition, and the second target intent knowledge output by the preset large language model is used as the target enhancement knowledge.

6. The method according to claim 4, characterized in that, The intent knowledge enhancement method includes: a first intent recognition method and a second intent recognition method, and the enhancement mode is further used to indicate the intent knowledge fusion strategy of the first intent recognition method and the second intent recognition method; The step of using the intent knowledge enhancement method to identify the intent recognition request and obtain target intent knowledge as the target enhancement knowledge includes: According to the first intent recognition method, a preset first natural language processing (NLP) model is used to recognize and process the intent recognition request, and the first target intent knowledge output by the first NLP model is obtained. According to the second intent recognition method, a preset intent recognition prompt word template is used to construct prompt words for the intent recognition request to obtain the corresponding intent recognition prompt words; Based on the intent recognition prompt words, a preset large language model is used to perform intent recognition, and the second target intent knowledge output by the preset large language model is obtained. The intent knowledge fusion strategy is used to fuse the first target intent knowledge and the second target intent knowledge to obtain the target enhancement knowledge.

7. The method according to claim 3, characterized in that, The knowledge definition description information of the knowledge to be enhanced indicates that the knowledge type of the knowledge to be enhanced is the slot knowledge type; The enhancement method is a slot knowledge enhancement method, and the enhanced knowledge recognition request is a slot recognition request; The step of using the enhancement method of the knowledge to be enhanced to perform knowledge enhancement recognition processing on the knowledge enhancement recognition request to obtain the target enhanced knowledge includes: The slot knowledge enhancement method is used to identify and process the slot recognition request to obtain target slot knowledge as the target enhanced knowledge.

8. The method according to claim 7, characterized in that, The slot knowledge enhancement method includes: a first slot recognition method and a second slot recognition method. The enhancement mode is also used to indicate the slot knowledge fusion strategy of the first slot recognition method and the second slot recognition method. The step of using the slot knowledge enhancement method to identify and process the slot recognition request to obtain target slot knowledge as the target enhanced knowledge includes: Based on the first slot recognition method, a preset second NLP model is used to recognize and process the slot recognition request, thereby obtaining the first target slot knowledge output by the second NLP model; According to the second slot recognition method, a preset slot recognition prompt word template is used to construct prompt words for the slot recognition request to obtain the corresponding slot recognition prompt words; Based on the slot recognition prompts, an intent recognition is performed using a preset large language model to obtain the second target slot knowledge output by the preset large language model. The target slot knowledge is obtained by fusing the first target slot knowledge and the second target slot knowledge using the slot knowledge fusion strategy.

9. A computer device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when the computer device is running, are executed by the processor to perform steps of a multi-turn dialogue method for a large-model-based intelligent dialogue system as described in any of claims 1 to 9.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, performs the steps of the multi-turn dialogue method of the large-model-based intelligent dialogue system as described in any of claims 1 to 9.

Images