Task-based dialogue system training method, task-based dialogue system deployment method, electronic equipment and storage medium

Abstract

The invention discloses a task-based dialogue system training method, a task-based dialogue system deployment method, electronic equipment and a storage medium, and the task-based dialogue system training method comprises the steps: obtaining task-based semantic data, and carrying out the pre-training of a large model, and obtaining a trained semantic expert model; cue words and data of at least four agents are obtained, and the agents comprise domain classification, slot analysis, COT semantics and fusion semantic scheduling; and performing fine adjustment on the semantic expert model by using the prompt words and data of the at least four agents to obtain a multi-agent semantic large model.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of large model technology, and in particular to a task-based dialogue system training and task-based dialogue system deployment method, electronic device and storage medium. Background Art

[0002] With the rapid development of artificial intelligence technology, large models have become a key force driving the evolution of dialogue systems. From early statistics-based machine learning models, to pre-trained language models, to today's large-scale pre-trained language models, dialogue systems have undergone profound changes. As the amount of data increases, the models are becoming larger and more powerful. Dialogue systems based on large models are the product of fusion and evolution. They have strong reasoning and external interaction capabilities. Through prompts, large models can infer various information in the dialogue context, including dialogue status, user status, and various semantic information. In addition, large models can interact with the external environment, including different knowledge sources and multimodal information, to provide timely, accurate, and reliable responses.

[0003] In the prior art, guidance information is mainly generated based on user instruction statements, and the guidance information includes: information of multiple pre-packaged candidate operation objects; based on the guidance information, at least one target operation object is determined from the multiple candidate operation objects, and the task corresponding to the user instruction statement is split into at least one execution step, and each execution step corresponds to a target operation object; for any execution step, the target operation object corresponding to any execution step is called to obtain the execution result of any execution step through the target operation object. The defect is that it completely uses a generative task-based dialogue architecture and cannot be integrated with traditional task-based dialogues. In addition, the prompt words for the large model are generated completely based on the user's input, and the reliability of the content generated by the large model cannot be controlled, and pluggable task-based dialogue processing cannot be achieved. Summary of the invention

[0004] Embodiments of the present invention provide a task-based dialogue system training and task-based dialogue system deployment method, an electronic device, and a storage medium, which are used to solve at least one of the above-mentioned technical problems.

[0005] In a first aspect, an embodiment of the present invention provides a method for training a task-based dialogue system, comprising: obtaining task-based semantic data to pre-train a large model to obtain a trained semantic expert model; obtaining prompt words and data of at least 4 intelligent agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; using the prompt words and data of the at least 4 intelligent agents to fine-tune the semantic expert model to obtain a multi-agent semantic large model.

[0006] In a second aspect, an embodiment of the present invention provides a method for deploying a task-based dialogue system, comprising: obtaining prompt words required by at least 4 types of intelligent agents, wherein the intelligent agents include domain classification, slot resolution, COT semantics and fusion semantic scheduling; introducing the prompt words required by the at least 4 types of intelligent agents into the multi-agent semantic big model to obtain 4 different types of big model agents; combining the architecture of the different big model agents with the architecture of the traditional task-based model to form a fused task-based dialogue system.

[0007] In a third aspect, an embodiment of the present invention 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 execute any one of the above-mentioned task-based dialogue system training and task-based dialogue system deployment methods of the present invention.

[0008] In a fourth aspect, an embodiment of the present invention provides a storage medium, in which one or more programs including execution instructions are stored, and 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 task-based dialogue system training and task-based dialogue system deployment methods of the present invention.

[0009] In a fifth aspect, an embodiment of the present invention further provides a computer program product, comprising a computer program stored on a storage medium, wherein the computer program comprises program instructions, and when the program instructions are executed by a computer, the computer executes any one of the above-mentioned task-based dialogue system training and task-based dialogue system deployment methods.

[0010] The method of the present application improves the flexibility and scalability of the system through modular design, which facilitates functional upgrades and maintenance. This design allows developers to quickly respond to changes in business needs and accelerate the integration and testing of new functions. Then, intelligent scheduling is achieved through the confidence assessment algorithm, which improves the system's ability to parse complex queries. This scheduling mechanism ensures that the system can provide the most appropriate parsing strategy when faced with diverse user queries. Furthermore, the rapid deployment mechanism of new fields significantly reduces the time and cost of going online and improves market response speed. This mechanism allows the system to quickly expand to new business areas to meet changing market needs. By utilizing the semantic understanding ability of LLM, combined with domain-specific prompt words and slot definitions, the parsing accuracy of complex queries is improved. This improvement in accuracy is achieved through a large amount of domain-specific data and sophisticated model tuning. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0012] Figure 1 A flowchart of a task-based dialogue system training method provided by one embodiment of the present invention; Figure 2 A flowchart of a task-based dialogue system deployment method provided by an embodiment of the present invention; Figure 3 A flowchart of a multi-semantic agent large model training process for a specific example of a task-based dialogue system training and task-based dialogue system deployment method provided by an embodiment of the present invention; Figure 4 A multi-semantic agent large model deployment flow chart of a specific example of a task-based dialogue system training and task-based dialogue system deployment method provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] Please refer to Figure 1 , which shows a flow chart of a task-based dialogue system training method provided by an embodiment of the present invention.

[0015] like Figure 1 As shown, in step 101, task-based semantic data is obtained to pre-train the large model to obtain a trained semantic expert model; In step 102, prompt words and data of at least four agents are obtained, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; In step 103, the semantic expert model is fine-tuned using the prompt words and data of the at least four agents to obtain a multi-agent semantic large model.

[0016] In this embodiment, for step 101, the task-based dialogue system training device obtains a large amount of task-based semantic data to pre-train the large model, wherein the task-based semantic data does not need to be in a fixed format, and a trained semantic expert model can be obtained after the training is completed.

[0017] Then, for step 102, the task-based dialogue system training device obtains the prompt words and data of at least 4 specified agents, wherein the agents include domain classification, slot parsing, COT (Chain-of-Thought) semantics and fusion semantic scheduling; for example, domain classification only requires the preparation of one full set of classification data, and slot parsing requires the preparation of slot definition prompt words and slot parsing results in multiple domains / skills. COT semantic parsing requires the definition of a full set of semantic definitions in the prompt words, and fusion semantic scheduling requires the preparation of one full set of binary classification data. Among them, in the AI ​​large model, the role of the prompt word (prompt, AI model prompt word) is mainly to prompt the AI ​​model with the context of input information and parameter information of the input model. COT is a technology used in the field of artificial intelligence and machine learning, especially in natural language processing tasks. COT aims to improve the model's understanding and reasoning ability by simulating the human thinking process. Its core idea is to decompose the complex reasoning process into a series of simple, linear steps, thereby improving the model's reasoning performance and explanation ability.

[0018] Finally, for step 103, the task-based dialogue system training device uses the prompt words and data of at least 4 agents to fine-tune the semantic expert model to obtain a multi-agent semantic large model, where an agent refers to an agent that can perceive the environment and take actions to achieve specific goals. It can be software, hardware or a system with autonomy, adaptability and interaction capabilities. The agent perceives changes in the environment (such as through sensors or data input), makes judgments and decisions based on the knowledge and algorithms it has learned, and then performs actions to affect the environment or achieve predetermined goals. Agents are widely used in the field of artificial intelligence, and are commonly found in automation systems, robots, virtual assistants, and game characters. The core is that they can learn autonomously and evolve continuously to better complete tasks and adapt to complex environments. .

[0019] The method of this embodiment obtains task-based semantic data to pre-train the large model to obtain a trained semantic expert model, and then uses prompt words and data of at least 4 agents to fine-tune the semantic expert model to obtain a multi-agent semantic large model, thereby improving the adaptability and accuracy of the model.

[0020] In some optional embodiments, the task-based dialogue system can quickly replace or upgrade any module in the task-based dialogue system through interface calls, wherein each module in the task-based dialogue system follows a unified input and output specification. For example, the task-based dialogue system adopts a highly modular pluggable design, allowing developers to flexibly replace or upgrade the skill classification and slot parsing modules in the system according to business needs. The system significantly improves the parsing ability of complex user queries by integrating traditional semantic understanding and semantic agents based on large language models (LLM). For example, the pluggable design is the core of the system's flexibility, which allows developers to quickly replace or upgrade any module in the system through simple interface calls. This design is based on a standardized plug-in architecture, and each module follows a unified input and output specification to ensure interoperability between modules. For example, if support for smart home control needs to be added, developers only need to develop a new module that meets the interface specification and insert it into the system without modifying the existing code. The pluggable design not only improves the maintenance efficiency of the system, but also provides great convenience for the integration of third-party services and custom functions, greatly expanding the potential application scope of the system.

[0021] In some optional embodiments, after fine-tuning the semantic expert model using the prompt words and data of the at least four agents to obtain a multi-agent semantic large model, the method further comprises: New domains are integrated by developers defining new domain-specific prompts and slots. For example, if you want to add a domain that handles user health consultations, developers only need to define the intent and slots of the health consultation domain, such as "symptoms", "medication", etc., and then add relevant examples and instructions to the prompts. In this way, even without a large amount of training data, the system can quickly understand and respond to health-related queries. This rapid launch mechanism greatly shortens the cycle from concept to practical application of new domains, allowing the system to quickly respond to market changes and user needs.

[0022] enter Output Play Music Results for the category - "Music" Play Myth Field classification results - "Film and TV; Music" Table 1 As shown in Table 1, in some optional embodiments, the domain classification agent of the agent is used to receive user queries and output the most likely domain classification, and supports multi-domain output.

[0023] enter Output Increase the air conditioner by 2 degrees [air conditioner: object][increase by 2 degrees: value]=>adjustment content=temperature; intent=body control Table 2 As shown in Table 2, the slot parsing agent is used to parse user queries in a specified domain and output aligned and non-aligned slot information and intent. For example, in the field of vehicle control, the slot parsing agent may recognize the output shown in Table 2.

[0024] enter Output Play Zhang San's Legend Skill classification results: 1. Film and TV 2. Music Skill slot analysis results: 1. Film and TV: [play: operation][Zhang San: filmmaker]'s [myth: film title] => intent=play film and TV 2. Music: [play: operation][Zhang San: singer name]'s [myth: song title] => intent=play music Table 3 As shown in Table 3, the COT semantic agent is used to combine domain classification and slot resolution to provide semantic understanding results. The agent can directly complete all the functions of secondary classification in traditional task-based semantics. By using the COT method, it first determines the skill to which the user query belongs, and then performs slot resolution within the skill.

[0025] Fusion semantic scheduling is used to select processing paths based on user queries to optimize user experience. The agent can determine whether to distribute the user's query to a task-based dialogue system based on a large language model agent or a task-based dialogue system based on traditional semantics. By determining the degree of colloquialism of the user's query, colloquial queries can be distributed to the large language model agent, and standard task-based queries can be distributed to traditional semantics. For example, when the input "open the window for me" is used, the "vehicle control agent" is dispatched, and when the input "open the window" is used, the traditional vehicle control skills are dispatched.

[0026] In some optional embodiments, the domain classification agent uses a unified prompt word to define the classification standards and examples for the entire domain, thereby ensuring the consistency and accuracy of the classification. In a specific example, the prompt word may include various domain classifications (for example, classification: vehicle control; example: input: "open the window", output: domain classification-"vehicle control"). The slot resolution agent customizes the prompt words for each domain, clarifies the definition of slots and intents, and adapts to the specific needs of different domains. The COT semantic agent combines the definitions of domain classification and slot resolution to provide semantic understanding; the semantic scheduling agent selects the processing agent for processing the user query based on the degree of colloquialism of the user query. For example, the most appropriate processing agent can be selected based on the degree of colloquialism of the user query, thereby realizing intelligent distribution.

[0027] Please refer to Figure 2 , which shows a flowchart of a task-based dialogue system deployment method provided by an embodiment of the present invention.

[0028] like Figure 2 As shown, in step 201, prompt words required by at least 4 types of agents are obtained, wherein the agents include domain classification, slot parsing, COT semantics and fusion semantic scheduling; In step 202, the prompt words required by the at least four types of agents are introduced into the multi-agent semantic large model to obtain four different types of large model agents; In step 203, the architecture of the different large model agents is combined with the architecture of the traditional task-based model to form a fused task-based dialogue system.

[0029] In this embodiment, for step 201, the task-based dialogue system deployment device takes prompt words required by at least 4 types of agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling. For example, the domain classification agent only needs 1 copy of prompt words to define all domains. Slot parsing prompt words need to be defined in each domain. COT semantic parsing agent needs 1 copy of fully defined prompt words. Fusion semantic scheduling only needs 1 copy of prompt words.

[0030] Then, for step 202, the task-based dialogue system deployment device introduces the prompt words required by at least four types of agents into the multi-agent semantic large model to obtain four different types of large model agents.

[0031] Finally, for step 203, the task-based dialogue system deployment device combines the architecture of different large-model agents with the architecture of traditional task-based models to form a fused task-based dialogue system. For example, the architecture of large-model semantic applications constructed by different semantic agents is different. For the domain classification agent, it is necessary to combine the slot parsing module in the traditional task-based semantic parsing to form a new secondary architecture to complete the task of semantic parsing. For the slot parsing agent, it is necessary to combine the domain classification module in the traditional task-based semantic parsing to form a new secondary architecture. For the COT semantic parsing agent, it can be used directly as a task-based semantic parsing. For the fused semantic scheduling agent, it is necessary to combine the COT semantic parsing agent with the traditional task-based semantic parsing to form a new fused semantic parsing architecture.

[0032] The method of this embodiment combines the architecture of different large-model intelligent agents with the architecture of traditional task-based models to form a fused task-based dialogue system, thereby improving the ability to parse complex user queries.

[0033] In some optional embodiments, the combining of the architecture of the different large model agents with the architecture of the traditional task-based model to form a fused task-based dialogue system includes: The task-based dialogue system deployment device combines the domain classification agent of the agent with the slot parsing module in the traditional task-based semantic parsing to form a fused secondary architecture; the task-based dialogue system deployment device combines the slot parsing agent of the agent with the domain classification module in the traditional task-based semantic parsing to form a fused secondary architecture; the task-based dialogue system deployment device combines the fusion semantic scheduling agent of the agent with the traditional task-based semantic parsing to form a fused semantic parsing architecture. This improves the parsing ability for complex user queries.

[0034] Please refer to Figure 3, which shows a flow chart of the multi-semantic agent large model training process of a specific example of a task-based dialogue system training and task-based dialogue system deployment method provided by an embodiment of the present invention.

[0035] like Figure 3 As shown, training process 1: pre-training model.

[0036] In this process, a large amount of task-oriented semantic data needs to be prepared to pre-train the large model. The data in this process does not need to follow a fixed format. After the training is completed, a semantic expert model can be obtained.

[0037] Training process 2: Multi-agent data preparation.

[0038] In this process, you need to prepare prompts and data for the specified four agents, namely domain classification, slot analysis, COT semantic analysis, and fusion semantic scheduling. Domain classification only requires one full set of classification data, and slot analysis requires slot definition prompts and slot analysis results in multiple domains / skills. COT semantic analysis requires full semantic definitions in the prompt. Fusion semantic scheduling only requires one full set of binary classification data.

[0039] Training process 3: Expert model fine-tuning training.

[0040] The semantic expert model obtained in step 1 is fine-tuned using the fine-tuning data prepared in step 2, and finally an LLM (Large Language Model) that supports multiple agent tasks is obtained.

[0041] Please refer to Figure 4 , which shows a multi-semantic agent large model deployment flow chart of a specific example of a task-based dialogue system training and task-based dialogue system deployment method provided by an embodiment of the present invention.

[0042] like Figure 4 As shown, deployment process 1: prompt preparation.

[0043] Prepare the prompts required by the four types of agents. The domain classification agent only needs one prompt that defines all domains. The slot resolution prompt needs to be defined in each domain. The COT semantic resolution agent needs one prompt that is fully defined. Fusion semantic scheduling only needs one prompt.

[0044] Deployment process 2: Agent construction.

[0045] The constructed prompts of different agents are introduced into the multi-agent semantic large model obtained by the final training, and 4 different large model agents are obtained.

[0046] Deployment process 3: semantic application construction.

[0047] The architecture of the large model semantic application built by different semantic agents is different. For the domain classification agent, it is necessary to combine the slot parsing module in the traditional task-based semantic parsing to form a new secondary architecture to complete the semantic parsing task. For the slot parsing agent, it is necessary to combine the domain classification module in the traditional task-based semantic parsing to form a new secondary architecture. For the COT semantic parsing agent, it can be used directly as a task-based semantic parsing. For the fusion semantic scheduling agent, it is necessary to combine the COT semantic parsing agent with the traditional task-based semantic parsing to form a new fusion semantic parsing architecture.

[0048] The training and deployment process includes two stages: pre-training and fine-tuning. In the pre-training stage, online data and semantic results are used to fine-tune the LLM to form an expert model. The purpose of this stage is to allow the model to learn common language patterns and domain-specific semantic features. For example, by analyzing a large number of user queries and corresponding intents, the model can learn that "play" is usually related to the "music" or "film" fields.

[0049] In the fine-tuning training phase, the fine-tuning data of the four agents are summarized and the expert LLM is further trained to improve the adaptability to specific tasks. In the deployment phase, according to business needs, the original modules can be replaced or mixed deployment can be carried out to achieve rapid launch of new fields. In addition, the system also supports the collection of high-frequency data through data backflow, and the backflow training to LLM after annotation, which further improves the analysis accuracy of new fields.

[0050] In some other embodiments, embodiments of the present invention further provide a non-volatile computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the task-based dialogue system training and task-based dialogue system deployment methods in any of the above method embodiments; As an implementation mode, the non-volatile computer storage medium of the present invention stores computer executable instructions, and the computer executable instructions are configured as follows: Obtain task-based semantic data to pre-train the large model and obtain a trained semantic expert model; Obtain prompt words and data for at least 4 agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; The semantic expert model is fine-tuned using the prompt words and data of the at least four agents to obtain a multi-agent semantic large model.

[0051] As another embodiment, the non-volatile computer storage medium of the present invention stores computer executable instructions, and the computer executable instructions are configured as follows: Obtain prompt words required by at least 4 types of agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; Introducing the prompt words required by the at least four types of agents into the multi-agent semantic large model to obtain four different types of large model agents; The architectures of the different large-model intelligent agents are combined with the architecture of the traditional task-based models to form a fused task-based dialogue system.

[0052] The non-volatile computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the task-based dialogue system training and task-based dialogue system deployment device, etc. In addition, the non-volatile computer-readable storage medium 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 non-volatile computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the task-based dialogue system training and task-based dialogue system deployment 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.

[0053] An embodiment of the present invention also provides 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 task-based dialogue system training and task-based dialogue system deployment methods.

[0054] Figure 5 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 5 As shown, the device includes: one or more processors 510 and a memory 520, Figure 5 A processor 510 is taken as an example. The device of the task-based dialogue system training and task-based dialogue system deployment method may also include: an input device 530 and an output device 540. 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 5The example of connecting via a bus is taken. The memory 520 is the above-mentioned non-volatile computer-readable storage medium. 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, the task-based dialogue system training and task-based dialogue system deployment methods of the above-mentioned method embodiments are implemented. The input device 530 can receive input digital or character information, and generate key signal input related to user settings and function control of the task-based dialogue system training and task-based dialogue system deployment devices. The output device 540 may include display devices such as display screens.

[0055] The above product can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0056] As an implementation mode, the electronic device is applied to a task-based dialogue system training and a task-based dialogue system deployment device, and includes: 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: Obtain task-based semantic data to pre-train the large model and obtain a trained semantic expert model; Obtain prompt words and data for at least 4 agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; The semantic expert model is fine-tuned using the prompt words and data of the at least four agents to obtain a multi-agent semantic large model.

[0057] As another implementation, the electronic device is applied to a task-based dialogue system training and a task-based dialogue system deployment device, including: 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: Obtain prompt words required by at least 4 types of agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; Introducing the prompt words required by the at least four types of agents into the multi-agent semantic large model to obtain four different types of large model agents; The architectures of the different large-model intelligent agents are combined with the architecture of the traditional task-based models to form a fused task-based dialogue system.

[0058] The electronic device of the embodiment of the present application exists in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by their mobile communication functions and their main purpose is to provide voice and data communications. These terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.

[0059] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, such as iPad.

[0060] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players (such as iPods), handheld game consoles, e-books, as well as smart toys and portable car navigation devices.

[0061] (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to the general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, manageability, etc.

[0062] (5) Other electronic devices with data interaction functions.

[0063] 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, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0064] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes 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 of each embodiment or some parts of the embodiment.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 invention.

Claims

1. A task-based dialogue system training method, comprising: Obtain task-based semantic data to pre-train the large model and obtain a trained semantic expert model; Obtain prompt words and data for at least 4 agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; The semantic expert model is fine-tuned using the prompt words and data of the at least four agents to obtain a multi-agent semantic large model.

2. The method according to claim 1, wherein: The task-based dialogue system can quickly replace or upgrade any module in the task-based dialogue system through interface calls, wherein each module in the task-based dialogue system complies with a unified input and output specification.

3. The method according to claim 1, wherein: After fine-tuning the semantic expert model using the prompt words and data of the at least four agents to obtain a multi-agent semantic large model, the method further comprises: Integrate new domains by defining new domain-specific prompts and slots by developers.

4. The method according to claim 1, wherein: The domain classification agent of the agent is used to receive user queries and output the most likely domain classification, and supports multi-domain output; The slot parsing agent is used to parse user queries in a specified domain and output aligned and non-aligned slot information and intent; COT semantic agent is used to combine domain classification and slot parsing to provide semantic understanding results; Fusion semantic scheduling is used to select processing paths based on user queries to optimize user experience.

5. The method according to claim 4, wherein: The domain classification agent uses a unified prompt word to define classification standards and examples for the entire domain; The slot parsing agent customizes prompt words for each field to clarify the definition of slots and intents; The COT semantic agent combines the definitions of domain classification and slot resolution to provide semantic understanding; The semantic scheduling agent selects a processing agent for processing the user query according to the colloquial level of the user query.

6. A method for deploying a task-based dialogue system, comprising: Obtain prompt words required by at least 4 types of agents, wherein the agents include domain classification, slot parsing, COT semantics, and fusion semantic scheduling; Introducing the prompt words required by the at least four types of agents into the multi-agent semantic large model to obtain four different types of large model agents; The architectures of the different large-model intelligent agents are combined with the architecture of the traditional task-based models to form a fused task-based dialogue system.

7. The method according to claim 6, wherein: The combining of the architecture of the different large model agents with the architecture of the traditional task-based model to form a fused task-based dialogue system includes: Combining the domain classification agent of the agent with the slot parsing module in the traditional task-based semantic parsing to form a fused secondary architecture; Combining the slot parsing agent of the agent with the domain classification module in the traditional task-based semantic parsing to form a fused secondary architecture; The fusion semantic scheduling agent of the agent is combined with the traditional task-based semantic parsing to form a fusion semantic parsing architecture.

8. 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 7.

9. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images