Multi-round dialogue sample data generation method, device, equipment and product

Abstract

The invention relates to the technical field of data processing, and particularly provides a multi-round dialogue sample data generation method, device, equipment and product, and the method can comprise the steps: inputting dialogue prompt information into a large language model, and generating system reply information which accords with a dialogue target in the dialogue prompt information; wherein the dialogue prompt information further comprises seed dialogue content; the system reply information comprises a system action, a system reply content and a system slot position in the system reply content; inputting the seed dialogue content into a pre-trained task-oriented dialogue model to obtain a model slot position in output content; based on the system slot position and the model slot position, determining a dialogue sample; the dialogue sample is used for training a dialogue model. According to the embodiment of the invention, high-quality dialogue samples can be automatically generated.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and more specifically, to a method, apparatus, device, and product for generating multi-round conversation sample data. Background Art

[0002] With the continuous development of artificial intelligence technology, the need for accurate conversations with users has become a common concern across all industries. Currently, multi-round conversations with users are usually achieved by deploying trained conversation models.

[0003] Currently, when training conversation models, historical conversation data from various scenarios is typically collected as sample datasets. However, in actual operations, the vast number of scenarios involved makes collecting historical conversation data inefficient and unable to guarantee sample quality.

[0004] Therefore, how to provide a technical solution for an efficient method of generating multi-round dialogue sample data has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The purpose of some embodiments of the present application is to provide a method, device, equipment and product for generating multi-round dialogue sample data. Through the technical solutions of the embodiments of the present application, efficient and high-quality generation of multi-round dialogue samples can be achieved, providing rich sample data for training dialogue models and improving model training accuracy.

[0006] In a first aspect, some embodiments of the present application provide a method for generating multi-round dialogue sample data, comprising: inputting dialogue prompt information into a large language model to generate system response information that meets the dialogue goal in the dialogue prompt information; wherein, the dialogue prompt information also includes: seed dialogue content; the system response information includes: system action, system response content and system slots in the system response content; inputting the seed dialogue content into a pre-trained task-oriented dialogue model to obtain model slots in the output content; determining dialogue samples based on the system slots and the model slots; wherein, the dialogue samples are used to train the dialogue model.

[0007] Some embodiments of this application first parse conversation prompts using a large language model to generate system responses; then, using a task-oriented conversation model, they process the seed conversation content to obtain model slots; and based on the model slots and the slots in the system response, they determine the output conversation samples. These embodiments of this application can automatically generate high-quality conversation samples through the model, improving the richness of the conversation samples, providing a high-quality training set for conversation model training, and improving the model's training accuracy.

[0008] In some embodiments, determining the dialogue sample based on the system slot and the model slot includes: if the system reply content does not contain the model slot, filling the model slot into the system reply information to obtain the dialogue sample in the first form; if the system slot and the model slot are of the same type and have different values, or the system slot and the model slot are of different types and have the same value, using the system reply information as the dialogue sample in the second form.

[0009] Some embodiments of the present application determine whether the system response content contains model slots and the relationship between the system slots and the model slots to determine the generated dialogue samples and improve the accuracy and quality of the dialogue samples.

[0010] In some embodiments, after determining the dialogue samples based on the system slots and the model slots, the method further includes: counting the slot detection situations in the dialogue samples of the first form and the second form to obtain statistical results; and improving the dialogue prompt information and the parsing rules within the large language model through the statistical results to facilitate obtaining a new round of the dialogue samples.

[0011] Some embodiments of the present application improve the parsing rules of dialogue prompt information and models by statistically analyzing the generation of dialogue samples, so as to generate high-quality dialogue samples.

[0012] In some embodiments, the method further includes: generating diversified dialogue samples by modifying the content of the dialogue prompt information.

[0013] Some embodiments of the present application can efficiently generate diverse dialogue samples in various scenarios by modifying dialogue prompt information, with high richness.

[0014] In some embodiments, before inputting the dialogue prompt information into the large language model, the method further includes: constructing seed data corresponding to different field combinations in multiple fields; wherein the seed data includes slot values ​​of multiple slot types; rewriting the description in the seed data and replacing the slot value according to a replacement rule to obtain new seed data; generating a random dialogue content combination and the dialogue target based on the new seed data; and selecting the seed dialogue content from the random dialogue content combination according to a seed selection rule.

[0015] Some embodiments of the present application provide reliable data support for subsequent conversation sample generation by constructing high-quality seed conversation content.

[0016] In some embodiments, the method further includes: based on the semantic coverage and semantic repetition of the conversation sample, changing the replacement rule and / or the seed selection rule to obtain diversified seed conversation content.

[0017] Some embodiments of the present application can generate richer seed conversation content by updating replacement rules and seed selection rules, and then obtain rich and high-quality conversation samples.

[0018] In some embodiments, the seed selection rule is slot overlap; selecting the seed conversation content from the random conversation content combination according to the seed selection rule includes: counting the slot overlap in the random conversation content; and using the conversation content with the highest slot overlap in the random conversation content combination as the seed conversation content.

[0019] Some embodiments of the present application can achieve high-quality selection of seed conversation content.

[0020] In a second aspect, some embodiments of the present application provide a device for generating multi-round dialogue sample data, including: a first generation module, used to input dialogue prompt information into a large language model, and generate system response information that meets the dialogue goal in the dialogue prompt information; wherein, the dialogue prompt information also includes: seed dialogue content; the system response information includes: system action, system response content and system slots in the system response content; a model processing module, used to input the seed dialogue content into a pre-trained task-oriented dialogue model to obtain model slots in the output content; a dialogue sample generation module, used to determine dialogue samples based on the system slots and the model slots; wherein, the dialogue samples are used to train the dialogue model.

[0021] In a third aspect, some embodiments of the present application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the method described in any embodiment of the first aspect.

[0022] In a fourth aspect, some embodiments of the present application provide an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor can implement a method as described in any embodiment of the first aspect when executing the program.

[0023] In a fifth aspect, some embodiments of the present application provide a computer program product, comprising a computer program, wherein the computer program, when executed by a processor, can implement the method described in any embodiment of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of some embodiments of the present application, the following is a brief introduction to the drawings required for use in some embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 System diagram for generating multi-turn conversation sample data provided by some embodiments of the present application; Figure 2 One of the flow charts of the method for generating multi-round dialogue sample data provided in some embodiments of the present application; Figure 3 Flowchart 2 of the method for generating multi-round conversation sample data provided in some embodiments of the present application; Figure 4 A block diagram of an apparatus for generating multi-round conversation sample data provided in some embodiments of the present application; Figure 5 A schematic diagram of an electronic device is provided for some embodiments of the present application. DETAILED DESCRIPTION

[0026] The technical solutions in some embodiments of the present application will be described below in conjunction with the drawings in some embodiments of the present application.

[0027] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0028] A large language model (LLM) is a deep learning model trained using large amounts of text data, enabling it to generate natural language text or understand the meaning of language text. These models can provide in-depth knowledge and language production on a variety of topics by training on large datasets. The core idea is to learn the patterns and structure of natural language through large-scale unsupervised training, simulating the human language cognition and generation process to a certain extent. To address the low efficiency and poor quality of conversation sample construction in existing technologies, the embodiments of the present application automatically generate conversation samples in different fields through a trained large language model, thereby providing rich corpus content for training the conversation model.

[0029] The following is combined with Figure 1The overall structure of the system for generating multi-round dialogue sample data provided by some embodiments of the present application is exemplified.

[0030] like Figure 1 As shown, some embodiments of the present application provide a system diagram for generating multi-round conversation sample data. The system for generating multi-round conversation sample data may include: a terminal 100 and a server 200. Terminal 100 may send content such as the application domain of a determined conversation sample and the slot type within that application domain to server 200. Server 200 may construct seed conversation content and conversation objectives based on the application domain, slot type, and other content. The seed conversation content and conversation objectives are then input into a pre-trained large language model, and system response information is output. Simultaneously, a pre-trained task-oriented conversation model is used to process the seed conversation content and output the corresponding model slot. Finally, based on the relationship between the system slot and the model slot in the system response information, the output conversation sample is determined.

[0031] In some embodiments of the present application, terminal 100 may be a mobile terminal or a non-portable computer terminal, which is not specifically limited in this embodiment of the present application. It is understood that the pre-trained large language model and task-oriented dialogue model are pre-deployed to server 200. The training process is the same as conventional training methods and will not be described in detail in this application.

[0032] The following is combined with Figure 2 The implementation process of generating multi-round dialogue sample data performed by the server 200 provided in some embodiments of the present application is exemplified.

[0033] Please see the attached Figure 2 , Figure 2 A flowchart of a method for generating multi-round conversation sample data is provided for some embodiments of the present application. The method for generating multi-round conversation sample data may include: S210: Input the conversation prompt information into a large language model to generate system response information that meets the conversation goal in the conversation prompt information. The conversation prompt information also includes seed conversation content, and the system response information includes a system action, system response content, and a system slot in the system response content. The large language model is an LLM model.

[0034] For example, in a specific embodiment of the present application, pre-constructed seed conversation content and conversation goal (as a specific example of a conversation goal) are input into the LLM model as prompt words (as a specific example of conversation prompt information) to generate a first round of conversation that meets the goal (as a specific example of system response information).

[0035] For example, the LLM model can select a model with chaining reasoning capabilities, such as qwen3. Using the seed conversation content and a goal, qwen3 is used to generate a conversation that meets the goal. The first round of user speech and the corresponding slots are generated. The LLM-generated results are then parsed and formatted. If the generated results include multiple rounds of conversation, only the results from the current round are used (with the ending sentence in the prompt as the split point). To balance generation quality and efficiency, chaining is removed during the first round of generation. If the generated format does not match, chaining is used again. This involves concatenating the current round's results and the slots within them into the prompt. This is then fed into the chaining-enabled LLM to generate system actions, system responses, and corresponding system slots (a specific example of a system slot). To balance generation quality and efficiency, chaining is removed during generation. If the generated format does not match, chaining is used again.

[0036] That is, when generating a conversation, each round of a conversation interaction generates a user speech, a slot within the user speech, a system action, a system speech, and a slot within the system speech. In each round, the user speaks first, followed by the system, until the conversation ends (i.e., "goodbye" appears in the system slot) or the maximum number of rounds (e.g., 20) is reached. This means that the LLM model may generate multiple rounds of conversation based on the input content, and the content related to the prompt word must be used as the first round of conversation.

[0037] In some embodiments of the present application, before executing S210, the method for generating multi-round dialogue sample data further includes: S201 , constructing seed data corresponding to different field combinations in multiple fields; wherein the seed data includes slot values ​​of multiple slot types.

[0038] For example, in a specific embodiment of this application, we first determine the slot types in a multi-domain schema and the multiple object instances for each slot. For example, the movie schema contains slot types such as movie title, theme, and director. The slot value (i.e., slot position value) is the content contained in each slot type, such as "watch movie" in the following example.

[0039] For example, a movie schema example: Name: Watching Movies Description: Provides movie search and movie watching services Slots: Movie name, movie theme, director, starring actor, rating operate: To search for a movie, you need one or more of the following slots: movie title, movie theme, director, and starring actor. Play movies, required slot: movie name Pause, Required Slots: None Continue, Required Slots: None Fast forward, required slot: time Rewind, required slot: Time Jump to, required slot: Time User actions: provide slot, request slot, confirm, make suggestion System actions: Make a request, deny, confirm, request a slot, request other.

[0040] Based on various combinations of schema structure relationships in multiple domains, samples are manually or automatically constructed, with each combination serving as a sample (i.e., seed data). Each sample includes conversations from different domain combinations (e.g., a sample between movies and music, which serves as a specific example of seed data). For example, the movie conversation in the following example is a piece of seed data: goal: movie, movie theme: Romance movie Multi-round dialogue: User: Can you find me a Romance movie to watch? Action = Provide Slot, Movie Theme = Romance, Operation = Find Movie System: How about After? It is one of the 5 movies that I have found, and has an average rating of 5.8. Action = Make a Suggestion, Movie Name = After, Rating = 5.80 Search results = 5 movies found User: What about the other movies? Action = Request Other System: I also found Pillow Talk that has a rating of 7.5. Action = Making a Suggestion, Movie Name = Pillow Talk, Rating = 7.50 User: Sure. Action = Confirm.

[0041] S202: rewrite the description in the seed data, and replace the slot value according to a replacement rule to obtain new seed data.

[0042] For example, in a specific embodiment of this application, the small number of samples constructed above are used as seed data. Based on the schema structure, slot values ​​are randomly replaced (as a specific example of a replacement rule). For example, the movie title "After" is replaced with "SpiderMan." Rephrase processing (i.e., rewriting a sentence using a different expression) is performed to form new samples (as a specific example of new seed data).

[0043] S203: Generate a random conversation content combination and the conversation target based on the new seed data.

[0044] For example, in a specific embodiment of this application, based on the multiple single-domain schemas identified above, a random number of combined schemas (as a specific example of random conversation content combination) are generated. Multiple slots are randomly selected from the combined schemas, and goals corresponding to these slots are generated. For example, in this example, the goal is genre = romance movie.

[0045] S204: Select the seed conversation content from the random conversation content combination according to a seed selection rule.

[0046] In some embodiments of the present application, the seed selection rule is slot overlap, and S204 may include: counting the slot overlap in the random conversation content; and taking the conversation content with the highest slot overlap in the random conversation content combination as the seed conversation content.

[0047] For example, in a specific embodiment of the present application, conversation examples with a high degree of slot overlap in the combined schema of the new conversation are selected as seed conversation content. That is, the selected seed conversation content all contains the same slot. Alternatively, conversation examples with a degree of overlap exceeding a set threshold are selected, but the present embodiment is not limited to this.

[0048] S220: Input the seed dialogue content into a pre-trained task-oriented dialogue model to obtain a model slot in the output content.

[0049] For example, in a specific embodiment of the present application, in order to reduce errors in LLM model generation (such as slot not being included in the system response content), a TOD mini-model (as a specific example of a task-oriented dialogue model) is trained using existing dialogue data. The TOD mini-model receives the seed dialogue content constructed above, generates slot information (as a specific example of a model slot) and a system response output by the mini-model (as a specific example of output content).

[0050] S230: Determine a dialogue sample based on the system slot and the model slot; wherein the dialogue sample is used to train a dialogue model.

[0051] For example, in a specific embodiment of the present application, the final dialogue sample is generated through the system slot corresponding to the LLM model and the slot information output by the above-mentioned small model.

[0052] In some embodiments of the present application, S230 may include: if the system reply content does not contain the model slot, filling the model slot into the system reply information to obtain the first form of the conversation sample; if the system slot and the model slot are of the same type and have different values, or the system slot and the model slot are of different types and have the same value, using the system reply information as the second form of the conversation sample.

[0053] For example, in a specific embodiment of this application, when the slot information generated by the small model is not included in the system response content generated by the LLM, the slot information is added to the system response information to obtain a conversation sample in this case. When the slots of the two models conflict (for example, they have the same slot type but different slot values, or the same slot value corresponds to different slot types), the system response information output by the LLM model is retained as the conversation sample.

[0054] In addition, as the amount of data continues to increase, the TOD small model can be retrained. When the data volume reaches the set requirements, it can be adjusted to retain the output content of the small model.

[0055] In some embodiments of the present application, in the process of generating conversation samples, the method for generating multi-round conversation sample data may also include: counting the slot detection situations in the conversation samples of the first form and the second form to obtain statistical results; and improving the conversation prompt information and the parsing rules within the large language model through the statistical results to facilitate obtaining a new round of the conversation samples.

[0056] For example, in the specific embodiments of this application, two methods are proposed for determining conversation samples: one is to incorporate slot information into the system response information, and the other is to retain the results output by the LLM model. By statistically analyzing the specific situations of these two methods, errors in slot generation, including false detections and missed detections, can be detected and counted. Based on the statistical results, the prompt words and the parsing rules within the model can be improved to generate a new round of conversation samples, thereby improving the accuracy of conversation sample generation.

[0057] In some embodiments of the present application, the method for generating multi-round dialogue sample data may further include: generating diversified dialogue samples by modifying the content of the dialogue prompt information.

[0058] For example, in a specific embodiment of the present application, in order to increase the richness of the generated dialogue samples, different roles can be assigned to the prompt words, and a randomly generated text context can be used as part of the prompt words (i.e., the content of the prompt words is modified), so that the LLM model can generate rich samples and ultimately generate dialogue samples with higher richness.

[0059] In some embodiments of the present application, the method for generating multi-round conversation sample data may further include: based on the semantic coverage and semantic repetition of the conversation samples, changing the replacement rules and / or the seed selection rules to obtain diversified seed conversation content.

[0060] For example, in a specific embodiment of this application, the generated conversation samples are visualized to examine their semantic coverage and semantic repetition, and at least one of the seed selection rules and slot replacement rules used in the seed conversation content construction phase is improved to enrich the seed conversation content and thereby obtain richer conversation samples. For example, the slot replacement rule is modified from random replacement to fixed replacement.

[0061] The following is combined with Figure 3 The specific process of generating multi-round dialogue sample data provided by some embodiments of the present application is exemplified.

[0062] Please see the attached Figure 3 , Figure 3 A flowchart of a method for generating multi-round conversation sample data is provided for some embodiments of the present application.

[0063] The above process is described below as an example.

[0064] S310, constructing seed data corresponding to different field combinations in multiple fields.

[0065] S320, rewrite the description in the seed data and randomly replace the slot value to obtain new seed data.

[0066] S330: Generate a random conversation content combination and a conversation target based on the new seed data.

[0067] S340: Select seed conversation content from the random conversation content combination according to a seed selection rule.

[0068] Among them, the seed selection rule is the highest slot overlap.

[0069] S350: Input the seed conversation content and conversation goal into the LLM model to generate system response information that meets the conversation goal.

[0070] S360: Input the seed conversation content into the pre-trained task-oriented conversation model to obtain the model slot in the output content.

[0071] S370: Determine the final generated dialogue sample based on the system slot and model slot in the system response information.

[0072] It is understandable that the specific implementation process of S310~S370 can refer to the method embodiment provided above. To avoid repetition, the detailed description is appropriately omitted here.

[0073] By using the method of generating multi-round dialogue data using LLM proposed in some embodiments of the present application, rapid iteration can be performed from scratch to construct high-quality and diverse multi-round dialogue samples, thereby improving the accuracy of dialogue model training.

[0074] Please refer to Figure 4 , Figure 4 The following is a block diagram illustrating the components of an apparatus for generating multi-turn conversation sample data, as provided in some embodiments of the present application. It should be understood that the apparatus for generating multi-turn conversation sample data corresponds to the aforementioned method embodiments and is capable of executing each step involved in the aforementioned method embodiments. The specific functions of the apparatus for generating multi-turn conversation sample data can be found in the description above, and a detailed description is omitted here to avoid repetition.

[0075] Figure 4 The device for generating multi-round dialogue sample data includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the device for generating multi-round dialogue sample data. The device for generating multi-round dialogue sample data includes: a first generation module 410, which is used to input dialogue prompt information into a large language model to generate system response information that meets the dialogue target in the dialogue prompt information; wherein, the dialogue prompt information also includes: seed dialogue content; the system response information includes: system action, system response content and system slot in the system response content; a model processing module 420, which is used to input the seed dialogue content into a pre-trained task-oriented dialogue model to obtain a model slot in the output content; a dialogue sample generation module 430, which is used to determine a dialogue sample based on the system slot and the model slot; wherein, the dialogue sample is used to train the dialogue model.

[0076] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method, and will not be described in detail here.

[0077] Some embodiments of the present application further provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the operations corresponding to any of the above methods provided in the above embodiments.

[0078] Some embodiments of the present application further provide a computer program product, which includes a computer program, wherein when the computer program is executed by a processor, it can implement the operations corresponding to any of the above methods provided in the above embodiments.

[0079] like Figure 5 As shown, some embodiments of the present application provide an electronic device 500, which includes: a memory 510, a processor 520, and a computer program stored in the memory 510 and executable on the processor 520, wherein the processor 520 can implement a method as described in any of the above embodiments when reading the program from the memory 510 through the bus 530 and executing the program.

[0080] Processor 520 can process digital signals and can include various computing architectures, such as a complex instruction set computer architecture, a reduced instruction set computer architecture, or an architecture that implements a combination of multiple instruction sets. In some examples, processor 520 can be a microprocessor.

[0081] The memory 510 can be used to store instructions executed by the processor 520 or data related to the execution of instructions. These instructions and / or data may include code for implementing some or all functions of one or more modules described in the embodiments of this application. The processor 520 of the embodiment of the present disclosure can be used to execute the instructions in the memory 510 to implement the method shown above. The memory 510 includes dynamic random access memory, static random access memory, flash memory, optical memory, or other memory known to those skilled in the art.

[0082] The foregoing is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included within the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures.

[0083] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0084] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

Claims

1. A method for generating multi-round dialogue sample data, characterized in that: include: Inputting the dialogue prompt information into the large language model to generate system response information that meets the dialogue goal in the dialogue prompt information; wherein the dialogue prompt information also includes: seed dialogue content; the system response information includes: system action, system response content, and system slot in the system response content; Inputting the seed conversation content into a pre-trained task-oriented conversation model to obtain a model slot in the output content; Based on the system slot and the model slot, a dialogue sample is determined; wherein the dialogue sample is used to train the dialogue model.

2. The method according to claim 1, wherein The determining of the dialogue sample based on the system slot and the model slot includes: If the system reply content does not contain the model slot, then fill the model slot into the system reply information to obtain the first form of the conversation sample; If the system slot and the model slot are of the same type but have different values, or if the system slot and the model slot are of different types but have the same value, the system reply information is used as the second form of the conversation sample.

3. The method according to claim 2, wherein After determining the dialogue sample based on the system slot and the model slot, the method further includes: Counting slot detection situations in the first form and the second form of dialogue samples to obtain statistical results; The dialogue prompt information and the parsing rules within the large language model are improved based on the statistical results, so as to obtain a new round of dialogue samples.

4. The method according to any one of claims 1 to 3, wherein The method further includes: generating diversified dialogue samples by modifying the content of the dialogue prompt information.

5. The method according to any one of claims 1 to 3, wherein Before inputting the dialogue prompt information into the large language model, the method further includes: Constructing seed data corresponding to different field combinations in multiple fields; wherein the seed data includes slot values ​​of multiple slot types; Rewrite the description in the seed data and replace the slot value according to the replacement rule to obtain new seed data; generating a random conversation content combination and the conversation goal based on the new seed data; The seed conversation content is selected from the random conversation content combination according to a seed selection rule.

6. The method according to claim 5, wherein The method further comprises: Based on the semantic coverage and semantic repetition of the conversation samples, the replacement rule and / or the seed selection rule are changed to obtain diversified seed conversation content.

7. The method according to claim 5, wherein The seed selection rule is slot overlap; and selecting the seed conversation content from the random conversation content combination according to the seed selection rule includes: Counting the slot overlap in the random conversation content; The conversation content with the highest slot overlap in the random conversation content combination is used as the seed conversation content.

8. A device for generating multi-round dialogue sample data, characterized in that: include: A first generation module is configured to input the dialogue prompt information into a large language model and generate system response information that meets the dialogue goal in the dialogue prompt information; wherein the dialogue prompt information also includes: seed dialogue content; the system response information includes: system action, system response content, and system slot in the system response content; A model processing module, configured to input the seed dialogue content into a pre-trained task-oriented dialogue model to obtain a model slot in the output content; A dialogue sample generation module is used to determine a dialogue sample based on the system slot and the model slot; wherein the dialogue sample is used to train the dialogue model.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program executes the method according to any one of claims 1 to 7 when the processor runs the computer program.

10. A computer program product, characterized in that The computer program product comprises a computer program, wherein the computer program is executed by a processor to perform the method according to any one of claims 1 to 7.

Images