A structured template prompting method for scalable dialog state correction

Abstract

This invention relates to the field of scalable dialogue state correction technology, specifically a structured template prompting method for scalable dialogue state correction. The method is implemented based on a scalable dialogue state correction system and includes the following steps: receiving full dialogue interaction data and initial dialogue state representation; standardizing the input data; constructing a multi-dimensional scalable structured template prompting system; and clarifying the template hierarchy, field definitions, and extension rules. This invention achieves scalability and self-updating of the template system by constructing a three-level structured template prompting system and formulating clear extension rules, adapting to dynamic changes in multi-domain dialogue scenarios. Through multi-dimensional feature extraction and quantized matching, it improves scenario adaptation accuracy, ensures the relevance of template subsets, and employs a multi-dimensional deviation quantification detection and hierarchical prompt generation strategy to accurately locate deviations and provide adaptive prompts, thereby improving the accuracy and efficiency of dialogue state correction.

Description

Technical Field

[0001] This invention relates to the field of scalable dialogue state correction technology, specifically a structured template prompting method for scalable dialogue state correction. Background Technology

[0002] Dialogue state correction, as a core component of natural language interaction systems, directly determines the accuracy and fluency of dialogue interaction.

[0003] However, in general, existing technologies for dialogue state correction often employ fixed template prompts or single-dimensional deviation detection methods, which suffer from poor template scalability and insufficient scenario adaptability, failing to meet the dynamic changing needs of multi-domain dialogue scenarios. Furthermore, existing methods lack a deviation quantification detection mechanism, making it difficult to accurately pinpoint the dimension and degree of deviation in the dialogue state, resulting in weakly targeted prompts and poor correction effects. In addition, existing template prompt systems lack dynamic iteration capabilities, failing to self-update based on dialogue correction effects and newly added corpus data. Long-term use leads to decreased correction accuracy and reduced adaptability, making it difficult to meet the practical application needs of scalable dialogue scenarios.

[0004] In summary, a structured template prompting method for scalable dialogue state correction is needed to address the aforementioned issues. Summary of the Invention

[0005] The purpose of this invention is to provide a structured template prompting method for scalable dialogue state correction, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention proposes a structured template prompt method for scalable dialogue state correction. The method is implemented based on a scalable dialogue state correction system and includes the following steps: S1. Receive the full data of the dialogue interaction and the initial state representation of the dialogue, standardize the input data, construct a multi-dimensional and scalable structured template prompt system, and clarify the hierarchical division, field definition and extension rules of the template; S2. Extract features from the standardized dialogue data, combine dialogue scene tags to complete the initial matching between dialogue features and the structured template prompt system, and generate a scene-adapted template subset; S3. Based on the template subset, the initial state of the dialogue is parsed, and the state elements are extracted and mapped to complete the multi-dimensional detection of the dialogue state deviation and generate a quantitative state deviation result. S4. Based on the state deviation quantification results, call the dynamic generation rules of the structured template prompt system to generate targeted hierarchical structured prompt statements and determine the timing and frequency of prompt triggering; S5. Push structured prompts to the dialogue processing engine and perform dialogue state correction operations, and collect dialogue state feedback data in real time during the correction process; S6. Verify the effect of dialogue state correction from multiple dimensions. If the verification fails, optimize the structured prompt statement based on the feedback data and re-execute the correction operation. If the verification passes, record the key parameters of the correction process. S7. Based on the verification results of the correction effect and the newly added data of industry dialogue data, the structured template prompt system is dynamically expanded and the rules are iterated to realize the self-updating and scalability of the template system.

[0007] Preferably, the implementation process of step S1 is as follows: The entire dialogue interaction data is cleaned, feature-annotated, and standardized in format to remove invalid and noisy interaction data. Construct a three-level structured template prompt system that includes a scene layer, an element layer, and a prompt layer. The scene layer divides different dialogue domain scenarios, the element layer defines the core dialogue state elements in each scenario, and the prompt layer designs the basic prompt statement templates for the corresponding state elements. Establish extension rules for the template system, and clarify the template access standards for new scenarios, the modification specifications for existing templates, and the constraints for adding and deleting fields; The initial state representation of the dialogue is converted into a standardized data format that matches the element layer of the template system, thus completing the format adaptation of the initial state.

[0008] Preferably, the implementation process of step S2 is as follows: The semantic feature extraction algorithm extracts scene features, intent features, entity features and contextual features from standardized dialogue data to generate a dialogue feature vector. Based on preset scene label matching rules, the matching degree between the dialogue feature vector and each scene in the scene layer of the structured template prompt system is calculated. The algorithm formula is as follows: ; In the formula, This represents the matching degree between dialogue features and the target scene, with a value range of [0,1]. A match of ≥0.8 is considered a strong match, while a match of 0.5≤0.8 is considered a strong match. <0.8 is considered a medium match. A match less than 0.5 is considered a weak match. For dialogue feature vectors, For the feature vector of the target scene, This is the scene association coefficient, with a value range of [0.9, 1.1]. It is used when the dialogue context contains scene-specific entities. =1.1, otherwise =0.9; Templates corresponding to scenarios with a matching degree of medium or higher are selected, a subset of templates adapted to the scenario is generated, and the matching degree is recalculated after feature supplementation for weakly matched scenarios.

[0009] Preferably, the implementation process of step S3 is as follows: Based on the element layer definition of the template subset, a set of state elements is extracted from the initial state of the dialogue, and each state element is mapped one by one to the standard elements of the template subset. Dialogue state deviation is detected from three dimensions: element completeness, element accuracy, and contextual consistency. The deviation values ​​for each dimension and the overall deviation value are calculated. The algorithm formula is as follows: ; In the formula, This is the overall deviation value, ranging from [0,1]. A deviation ≥0.7 is considered a severe deviation, and 0.3 ≤ A deviation of <0.7 is considered a moderate deviation. A value <0.3 is considered a slight deviation. This represents the deviation value for element integrity. This represents the deviation value of the element accuracy. These are context consistency deviation values, and all three values ​​range from [0,1]. Based on the overall deviation value and the deviation values ​​of each dimension, a quantitative state deviation result is generated, which includes the deviation level, deviation dimension, and deviation element.

[0010] Preferably, the implementation process of step S4 is as follows: Establish a correspondence between status deviation levels and structured prompt statement generation rules: severe deviations are matched with all-element hierarchical prompt rules, moderate deviations are matched with core element prompt rules, and minor deviations are matched with single-element correction prompt rules. Following the template specifications of the prompt layer, and combining the deviation dimensions and deviation elements in the quantitative state deviation results, hierarchical structured prompt statements are generated. The statements include guidance on deviation element correction, contextual prompts, and intent confirmation content. Based on the real-time rhythm of the dialogue interaction, the timing of triggering prompts is determined. Severe deviations trigger multiple rounds of prompts at once, moderate deviations trigger two rounds of prompts in stages, and mild deviations trigger a single prompt. An upper limit threshold for the frequency of prompts is set.

[0011] Preferably, the implementation process of step S6 is as follows: The correction effect was verified from three dimensions: state correction completion, dialogue intent matching, and interaction fluency. Each dimension had a valid threshold. If any dimension fails to reach the qualified threshold, the correction effect verification is deemed unsuccessful. Based on the collected feedback data, the problem points of the prompt statement are located, the expression form and element guidance method of the statement are optimized, the structured prompt statement is regenerated, and the correction operation is performed. If all dimensions meet the acceptable threshold, the verification is considered successful. Record key parameters such as deviation level, prompt rules, trigger parameters, and correction time during this correction process, and complete the classification and storage of parameters.

[0012] Preferably, the implementation process of step S7 is as follows: Set the trigger conditions for template system expansion and rule iteration. Triggering is triggered when any one of the following three conditions is met: the cumulative number of correction verifications passed in a single scenario reaches a preset number, a new dialogue domain scenario is added, or the amount of new data in the industry dialogue corpus reaches a threshold. For new scenarios, complete the template construction and access according to the extension rules in step S1; for existing templates, optimize the element definition and prompt statements based on the key parameters of the correction process. Based on the newly added corpus data, the scene feature vector, matching degree calculation rules, and deviation value calculation rules are iteratively updated. The iterated rules and templates are then synchronized to the structured template prompting system to achieve self-updating and scalability of the system.

[0013] Based on the method, this invention also proposes an scalable dialogue state correction system for a structured template prompting method for scalable dialogue state correction, the system comprising: A data receiving and processing module, which uses standardized processing algorithms and system construction rules to execute the operations in step S1; The feature matching module uses a feature extraction algorithm and matching degree calculation rules to perform the operation in step S2.

[0014] Preferably, the system further includes: The deviation detection module is used to perform the operation in step S3 by using element mapping rules and deviation value calculation algorithms; The prompt generation module dynamically generates rules and triggering strategies to execute the operations in step S4.

[0015] Preferably, the system further includes: The correction and verification module uses feedback data collection rules and effect verification dimensions to execute the operations in steps S5 and S6. The system iteration module executes the operations in step S7 by extending the triggering conditions and rule update algorithm.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing a three-level structured template prompting system and formulating clear extension rules, this invention achieves the scalability and self-updating of the template system, adapting to the dynamic changes of multi-domain dialogue scenarios. Through multi-dimensional feature extraction and quantitative matching, it improves the scene adaptation accuracy and ensures the relevance of template subsets. By adopting a multi-dimensional deviation quantitative detection and hierarchical prompt generation strategy, it accurately locates deviations and provides adaptive prompts, improving the accuracy and efficiency of dialogue state correction. Through correction effect verification and rule iteration closed loop, it achieves dynamic optimization of the prompting system and algorithm rules, continuously ensuring correction accuracy. The various modules of the system work together to achieve full-process automation from data processing, deviation detection, correction verification, and system iteration, reducing the cost of manual intervention and significantly improving the stability and practicality of state correction in scalable dialogue scenarios. Attached Figure Description

[0017] Figure 1 A topology diagram of the scalable dialogue state correction system of the present invention is shown. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] For examples, please refer to Figure 1 This invention proposes a structured template prompt method for scalable dialogue state correction. The method is implemented based on a scalable dialogue state correction system, which includes: A data receiving and processing module, which uses standardized processing algorithms and system construction rules to execute the operations in step S1; The feature matching module uses a feature extraction algorithm and matching degree calculation rules to perform the operation in step S2.

[0020] In one specific embodiment, it should also be noted that the system further includes: The deviation detection module is used to perform the operation in step S3 by using element mapping rules and deviation value calculation algorithms; The prompt generation module dynamically generates rules and triggering strategies to execute the operations in step S4.

[0021] In one specific embodiment, it should also be noted that the system further includes: The correction and verification module uses feedback data collection rules and effect verification dimensions to execute the operations in steps S5 and S6. The system iteration module executes the operations in step S7 by extending the triggering conditions and rule update algorithm.

[0022] Based on the system, in practical applications, a structured template prompt method for scalable dialogue state correction specifically includes the following steps: S1. Receive the full data of the dialogue interaction and the initial state representation of the dialogue, standardize the input data, construct a multi-dimensional and scalable structured template prompt system, and clarify the hierarchical division, field definition and extension rules of the template; In a specific embodiment, it should also be noted that the implementation process of step S1 is as follows: The entire dialogue interaction data is cleaned, feature-annotated, and standardized in format to remove invalid and noisy interaction data. Construct a three-level structured template prompt system that includes a scene layer, an element layer, and a prompt layer. The scene layer divides different dialogue domain scenarios, the element layer defines the core dialogue state elements in each scenario, and the prompt layer designs the basic prompt statement templates for the corresponding state elements. Establish extension rules for the template system, and clarify the template access standards for new scenarios, the modification specifications for existing templates, and the constraints for adding and deleting fields; The initial state representation of the dialogue is converted into a standardized data format that matches the element layer of the template system, thus completing the format adaptation of the initial state.

[0023] S2. Extract features from the standardized dialogue data, combine dialogue scene tags to complete the initial matching between dialogue features and the structured template prompt system, and generate a scene-adapted template subset; In a specific embodiment, it should also be noted that the implementation process of step S2 is as follows: The semantic feature extraction algorithm extracts scene features, intent features, entity features and contextual features from standardized dialogue data to generate a dialogue feature vector. Based on preset scene label matching rules, the matching degree between the dialogue feature vector and each scene in the scene layer of the structured template prompt system is calculated. The algorithm formula is as follows: ; In the formula, This represents the matching degree between dialogue features and the target scene, with a value range of [0,1]. A match of ≥0.8 is considered a strong match, while a match of 0.5≤0.8 is considered a strong match. <0.8 is considered a medium match. A match less than 0.5 is considered a weak match. For dialogue feature vectors, For the feature vector of the target scene, This is the scene association coefficient, with a value range of [0.9, 1.1]. It is used when the dialogue context contains scene-specific entities. =1.1, otherwise =0.9; Templates corresponding to scenarios with a matching degree of medium or higher are selected, a subset of templates adapted to the scenario is generated, and the matching degree is recalculated after feature supplementation for weakly matched scenarios.

[0024] S3. Based on the template subset, the initial state of the dialogue is parsed, and the state elements are extracted and mapped to complete the multi-dimensional detection of the dialogue state deviation and generate a quantitative state deviation result. In a specific embodiment, it should also be noted that the implementation process of step S3 is as follows: Based on the element layer definition of the template subset, a set of state elements is extracted from the initial state of the dialogue, and each state element is mapped one by one to the standard elements of the template subset. Dialogue state deviation is detected from three dimensions: element completeness, element accuracy, and contextual consistency. The deviation values ​​for each dimension and the overall deviation value are calculated. The algorithm formula is as follows: ; In the formula, This is the overall deviation value, ranging from [0,1]. A deviation ≥0.7 is considered a severe deviation, and 0.3 ≤ A deviation of <0.7 is considered a moderate deviation. A value <0.3 is considered a slight deviation. This represents the deviation value for element integrity. This represents the deviation value of the element accuracy. These are context consistency deviation values, and all three values ​​range from [0,1]. Based on the overall deviation value and the deviation values ​​of each dimension, a quantitative state deviation result is generated, which includes the deviation level, deviation dimension, and deviation element.

[0025] S4. Based on the state deviation quantification results, call the dynamic generation rules of the structured template prompt system to generate targeted hierarchical structured prompt statements and determine the timing and frequency of prompt triggering; In a specific embodiment, it should also be noted that the implementation process of step S4 is as follows: Establish a correspondence between status deviation levels and structured prompt statement generation rules: severe deviations are matched with all-element hierarchical prompt rules, moderate deviations are matched with core element prompt rules, and minor deviations are matched with single-element correction prompt rules. Following the template specifications of the prompt layer, and combining the deviation dimensions and deviation elements in the quantitative state deviation results, hierarchical structured prompt statements are generated. The statements include guidance on deviation element correction, contextual prompts, and intent confirmation content. Based on the real-time rhythm of the dialogue interaction, the timing of triggering prompts is determined. Severe deviations trigger multiple rounds of prompts at once, moderate deviations trigger two rounds of prompts in stages, and mild deviations trigger a single prompt. An upper limit threshold for the frequency of prompts is set.

[0026] S5. Push structured prompts to the dialogue processing engine and perform dialogue state correction operations, and collect dialogue state feedback data in real time during the correction process; S6. Verify the effect of dialogue state correction from multiple dimensions. If the verification fails, optimize the structured prompt statement based on the feedback data and re-execute the correction operation. If the verification passes, record the key parameters of the correction process. In a specific embodiment, it should also be noted that the implementation process of step S6 is as follows: The correction effect was verified from three dimensions: state correction completion, dialogue intent matching, and interaction fluency. Each dimension had a valid threshold. If any dimension fails to reach the qualified threshold, the correction effect verification is deemed unsuccessful. Based on the collected feedback data, the problem points of the prompt statement are located, the expression form and element guidance method of the statement are optimized, the structured prompt statement is regenerated, and the correction operation is performed. If all dimensions meet the acceptable threshold, the verification is considered successful. Record key parameters such as deviation level, prompt rules, trigger parameters, and correction time during this correction process, and complete the classification and storage of parameters.

[0027] S7. Based on the verification results of the correction effect and the newly added data of industry dialogue data, the structured template prompt system is dynamically expanded and the rules are iterated to realize the self-updating and scalability of the template system.

[0028] In a specific embodiment, it should also be noted that the implementation process of step S7 is as follows: Set the trigger conditions for template system expansion and rule iteration. Triggering is triggered when any one of the following three conditions is met: the cumulative number of correction verifications passed in a single scenario reaches a preset number, a new dialogue domain scenario is added, or the amount of new data in the industry dialogue corpus reaches a threshold. For new scenarios, complete the template construction and access according to the extension rules in step S1; for existing templates, optimize the element definition and prompt statements based on the key parameters of the correction process. Based on the newly added corpus data, the scene feature vector, matching degree calculation rules, and deviation value calculation rules are iteratively updated. The iterated rules and templates are then synchronized to the structured template prompting system to achieve self-updating and scalability of the system.

[0029] Through the above steps, this invention constructs a three-level structured template prompting system and formulates clear extension rules, thereby achieving the scalability and self-updating of the template system and adapting to the dynamic changes of multi-domain dialogue scenarios. By extracting features from multiple dimensions and using quantitative matching, the accuracy of scene adaptation is improved, ensuring the relevance of the template subset; By employing a multi-dimensional deviation quantification detection and hierarchical prompt generation strategy, deviations are accurately located and adaptive prompts are provided, improving the accuracy and efficiency of dialogue state correction. By verifying the correction effect and implementing a closed-loop rule iteration, the prompting system and algorithm rules are dynamically optimized, continuously ensuring the correction accuracy. The system's various modules work together to automate the entire process from data processing and deviation detection to correction verification and system iteration, reducing the cost of manual intervention and significantly improving the stability and practicality of state correction in scalable dialogue scenarios.

[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A structured template prompt method for scalable dialogue state correction, the method being implemented based on a scalable dialogue state correction system, characterized in that, Includes the following steps: S1. Receive the full data of the dialogue interaction and the initial state representation of the dialogue, standardize the input data, construct a multi-dimensional and scalable structured template prompt system, and clarify the hierarchical division, field definition and extension rules of the template; S2. Extract features from the standardized dialogue data, combine dialogue scene tags to complete the initial matching between dialogue features and the structured template prompt system, and generate a scene-adapted template subset; S3. Based on the template subset, the initial state of the dialogue is parsed, and the state elements are extracted and mapped to complete the multi-dimensional detection of the dialogue state deviation and generate a quantitative state deviation result. S4. Based on the state deviation quantification results, call the dynamic generation rules of the structured template prompt system to generate targeted hierarchical structured prompt statements and determine the timing and frequency of prompt triggering; S5. Push structured prompts to the dialogue processing engine and perform dialogue state correction operations, and collect dialogue state feedback data in real time during the correction process; S6. Verify the effect of dialogue state correction from multiple dimensions. If the verification fails, optimize the structured prompt statement based on the feedback data and re-execute the correction operation. If the verification passes, record the key parameters of the correction process. S7. Based on the verification results of the correction effect and the newly added data of industry dialogue data, the structured template prompt system is dynamically expanded and the rules are iterated to realize the self-updating and scalability of the template system.

2. The structured template prompting method for scalable dialogue state correction according to claim 1, characterized in that, The implementation process of step S1 is as follows: The entire dialogue interaction data is cleaned, feature-annotated, and standardized in format to remove invalid and noisy interaction data. Construct a three-level structured template prompt system that includes a scene layer, an element layer, and a prompt layer. The scene layer divides different dialogue domain scenarios, the element layer defines the core dialogue state elements in each scenario, and the prompt layer designs the basic prompt statement templates for the corresponding state elements. Establish extension rules for the template system, and clarify the template access standards for new scenarios, the modification specifications for existing templates, and the constraints for adding and deleting fields; The initial state representation of the dialogue is converted into a standardized data format that matches the element layer of the template system, thus completing the format adaptation of the initial state.

3. The structured template prompting method for scalable dialogue state correction according to claim 2, characterized in that, The implementation process of step S2 is as follows: The semantic feature extraction algorithm extracts scene features, intent features, entity features and contextual features from standardized dialogue data to generate a dialogue feature vector. Based on preset scene label matching rules, the matching degree between the dialogue feature vector and each scene in the scene layer of the structured template prompt system is calculated. The algorithm formula is as follows: ； In the formula, This represents the matching degree between dialogue features and the target scene, with a value range of [0,1]. A match of ≥0.8 is considered a strong match, while a match of 0.5≤0.8 is considered a strong match. <0.8 is considered a medium match. A match less than 0.5 is considered a weak match. For dialogue feature vectors, For the feature vector of the target scene, This is the scene association coefficient, with a value range of [0.9, 1.1]. It is used when the dialogue context contains scene-specific entities. =1.1, otherwise =0.9; Templates corresponding to scenarios with a matching degree of medium or higher are selected, a subset of templates adapted to the scenario is generated, and the matching degree is recalculated after feature supplementation for weakly matched scenarios.

4. The structured template prompting method for scalable dialogue state correction according to claim 3, characterized in that, The implementation process of step S3 is as follows: Based on the element layer definition of the template subset, a set of state elements is extracted from the initial state of the dialogue, and each state element is mapped one by one to the standard elements of the template subset. Dialogue state deviation is detected from three dimensions: element completeness, element accuracy, and contextual consistency. The deviation values ​​for each dimension and the overall deviation value are calculated. The algorithm formula is as follows: ； In the formula, This is the overall deviation value, ranging from [0,1]. A deviation ≥0.7 is considered a severe deviation, and 0.3 ≤ A deviation of <0.7 is considered a moderate deviation. A value <0.3 is considered a slight deviation. This represents the deviation value for element integrity. This represents the deviation value of the element accuracy. These are context consistency deviation values, and all three values ​​range from [0,1]. Based on the overall deviation value and the deviation values ​​of each dimension, a quantitative state deviation result is generated, which includes the deviation level, deviation dimension, and deviation element.

5. A structured template prompting method for scalable dialogue state correction according to claim 4, characterized in that, The implementation process of step S4 is as follows: Establish a correspondence between status deviation levels and structured prompt statement generation rules: severe deviations are matched with all-element hierarchical prompt rules, moderate deviations are matched with core element prompt rules, and minor deviations are matched with single-element correction prompt rules. Following the template specifications of the prompt layer, and combining the deviation dimensions and deviation elements in the quantitative state deviation results, hierarchical structured prompt statements are generated. The statements include guidance on deviation element correction, contextual prompts, and intent confirmation content. Based on the real-time rhythm of the dialogue interaction, the timing of triggering prompts is determined. Severe deviations trigger multiple rounds of prompts at once, moderate deviations trigger two rounds of prompts in stages, and mild deviations trigger a single prompt. An upper limit threshold for the frequency of prompts is set.

6. A structured template prompting method for scalable dialogue state correction according to claim 5, characterized in that, The implementation process of step S6 is as follows: The correction effect was verified from three dimensions: state correction completion, dialogue intent matching, and interaction fluency. Each dimension had a valid threshold. If any dimension fails to reach the qualified threshold, the correction effect verification is deemed unsuccessful. Based on the collected feedback data, the problem points of the prompt statement are located, the expression form and element guidance method of the statement are optimized, the structured prompt statement is regenerated, and the correction operation is performed. If all dimensions meet the acceptable threshold, the verification is considered successful. Record key parameters such as deviation level, prompt rules, trigger parameters, and correction time during this correction process, and complete the classification and storage of parameters.

7. A structured template prompting method for scalable dialogue state correction according to claim 6, characterized in that, The implementation process of step S7 is as follows: Set the trigger conditions for template system expansion and rule iteration. Triggering is triggered when any one of the following three conditions is met: the cumulative number of correction verifications passed in a single scenario reaches a preset number, a new dialogue domain scenario is added, or the amount of new data in the industry dialogue corpus reaches a threshold. For new scenarios, complete the template construction and access according to the extension rules in step S1; for existing templates, optimize the element definition and prompt statements based on the key parameters of the correction process. Based on the newly added corpus data, the scene feature vector, matching degree calculation rules, and deviation value calculation rules are iteratively updated. The iterated rules and templates are then synchronized to the structured template prompting system to achieve self-updating and scalability of the system.

8. A scalable dialogue state correction system applied to the structured template prompting method for scalable dialogue state correction as described in any one of claims 1-7, characterized in that, The system includes: A data receiving and processing module, which uses standardized processing algorithms and system construction rules to execute the operations in step S1; The feature matching module uses a feature extraction algorithm and matching degree calculation rules to perform the operation in step S2.

9. The scalable dialogue state correction system according to claim 8, characterized in that, The system also includes: The deviation detection module is used to perform the operation in step S3 by using element mapping rules and deviation value calculation algorithms; The prompt generation module dynamically generates rules and triggering strategies to execute the operations in step S4.

10. The scalable dialogue state correction system according to claim 8, characterized in that, The system also includes: The correction and verification module uses feedback data collection rules and effect verification dimensions to execute the operations in steps S5 and S6. The system iteration module executes the operations in step S7 by extending the triggering conditions and rule update algorithm.

Images